The Miracle of Microfinance? Evidence from a Randomized Evaluation

• p. 5 Introduction, last line

  was not significantly different (38.5% in

• p. 6 Introduction, first line

  treatment and 33% in control)

• p. 15 Section 2.4

  33.1% of households had borrowed from an MFI in the former control slums, and 33.7% in the treatment slums

• p. 45 Table 3 Panel B col 1

  0.0058

Only the authors can say what is right here.

Judge's note: Table 3 Panel B col 1 (p45 upright) gives control mean 0.33 and treatment coefficient 0.0058 (0.030), not significant, which fits 33.7% vs 33.1%. The 38.5% (intro, p5) and 42% (fn 30, p35) treatment figures cannot be squared with that table, and a 5.5 pp gap does not sit easily with the 'not significantly different' statement in the same sentence. The conclusion text gives an overall 38%. No alternative definition is stated anywhere. Footnote 30 also prints a redundant '% percent'.

Second check: another model agreed

Report a problem with this finding

This finding is

wrong right unclear or overstated already corrected (erratum, later version)

I am an author of this paper Send

• p. 10 Section 2.2

  at 89.9% in the treatment group and 90.2% in the control group

• p. 55 Table A2 Panel A

  0.8889

• p. 10 Section 2.2

  89.9% in the treatment group and 90.2% in the control group

• p. 55 Table A2 Panel A

  Found in endline 2, in treatment 0.8889

Judge's note: Checked page 10 (printed p.8) and Table A2 on page 55. The control figure, 0.9017, matches the 90.2% in the text. The treatment figure, 0.8889, rounds to 88.9%, not 89.9%. Both refer to the same quantity, so rounding cannot account for the difference.

Second check: another model agreed

Report a problem with this finding

This finding is

wrong right unclear or overstated already corrected (erratum, later version)

I am an author of this paper Send

• p. 12 Growth between 2005 and 2010

  Average household consumption rose from Rs. 5,485 to Rs. 7,662

• p. 43 Table 1, Household expenditure

  5,485

• p. 12 Section 2.3

  Average household consumption rose from Rs. 5,485

• p. 12 Section 2.3 Growth

  Average household consumption rose from Rs. 5,485 to Rs. 7,662 in 2007

Judge's note: Table 1 (page 43) gives Household expenditure as 5,485 (Treatment) and 5,208 (Control). The pooled mean is 'just under Rs. 5350' (p.9). Elsewhere in this growth paragraph the baseline figures are control values (savings 34% = 0.34, loans 68%, businesses 24%), so 5,485 does not fit.

Second check: another model agreed

Report a problem with this finding

This finding is

wrong right unclear or overstated already corrected (erratum, later version)

I am an author of this paper Send

• p. 12 Growth section

  between baseline and EL2 (68% versus 56%)

• p. 44 Table 2, Waterproof house

  0.558

• p. 44 Table 2

  Waterproof house 0.558 0.679

• p. 12 Section 2.3 Growth

  12 percentage point increase in the likelihood the family’s house was waterproof between baseline and EL2 (68% versus 56%)

Judge's note: Table 2 (page 44) gives Waterproof house as 0.558 (EL1) and 0.679 (EL2), a difference of 0.121. Table 1 reports no baseline waterproof figure. The comparison shown is EL1 to EL2, not baseline to EL2.

Second check: another model agreed

Report a problem with this finding

This finding is

wrong right unclear or overstated already corrected (erratum, later version)

I am an author of this paper Send

• p. 12 Section 2.3

  Almost none (0.03%) had any health insurance.

• p. 43 Table 1, Household with health insurance

  0.003

• p. 12 Section 2.3

  Almost none (0.03%) had any health insurance

• p. 43 Table 1

  Household with health insurance 0.003 0.003

Judge's note: Table 1 (page 43) shows 0.003 for both treatment and control. As a percentage that is 0.3%, so the text figure is ten times too small.

Second check: another model agreed

Report a problem with this finding

This finding is

wrong right unclear or overstated already corrected (erratum, later version)

I am an author of this paper Send

• p. 12 Section 2.3

  spending Rs. 550 ($54) or more on a health shock

• p. 43 Table 1

  Household spent Rs. 500 on health shock, previous year

• p. 13 Growth section

  spending Rs. 500 or more on a health shock

• p. 12 Section 2.3

  spending Rs. 550 ($54) or more

Only the authors can say what is right here.

Judge's note: The same variable has a Rs. 500 threshold in Table 1, Table 2 and the p.11 text. The $54 is consistent with Rs. 550 at the paper's implied PPP rate of about 10 Rs per dollar (9,900 to $980), so reviewer note 4's 9.2 rate does not hold. Only the authors can say which threshold applies at baseline.

Second check: another model agreed

Report a problem with this finding

This finding is

wrong right unclear or overstated already corrected (erratum, later version)

I am an author of this paper Send

• p. 13 Growth section

  50% at baseline to 53% at EL1 and 45% at EL2

• p. 12 Section 2.3

  50% of households who had a sick member had to borrow

• p. 43 Table 1 Panel B

  0.581

Only the authors can say what is right here.

Judge's note: Table 1 Panel B (page 43): treatment 0.5, control 0.581. Table 2 gives 0.530 and 0.454, matching the EL1 and EL2 figures. The growth paragraph otherwise uses control baseline values. A pooled value would not round to 50% unless control households carried under about 6% of the weight. With 58% as the baseline the trend reads differently, so the authors should confirm.

Second check: another model agreed

Report a problem with this finding

This finding is

wrong right unclear or overstated already corrected (erratum, later version)

I am an author of this paper Send

• p. 14 Section 2.4

  Attanasio et al. (2011) find a much larger increase, 48 percentage points

• p. 32 Section 5

  57 percentage points higher in AADFH

• p. 53 Table 10, Mongolia col 2

  increase of 48% in the probability of receiving microcredit

• p. 14 Section 2.4

  find a much larger increase, 48 percentage points

Only the authors can say what is right here.

Judge's note: Page 32 (printed p.30) describes 'Access to microfinance from the MFI who participated in the trials' and says the group-liability results are reported. That is the same quantity as on p.12 and in Table 10 col 2 (page 53, upright), which give 48. Nothing in the paper defines 57 as a different measure.

Second check: another model agreed

Report a problem with this finding

This finding is

wrong right unclear or overstated already corrected (erratum, later version)

I am an author of this paper Send

• p. 35 Conclusion and footnote 30

  The takeup rate is 42% in treatment areas and 33% percent in control areas.

• p. 35 Conclusion text

  38% of households borrow from an MFI

• p. 15 Section 2.4

  and 33.7% in the treatment slums

• p. 35 Conclusion

  only 38% of households borrow from an MFI

Only the authors can say what is right here.

Judge's note: Checked Table 3 Panel B col 1 (upright p.45): 0.0058 (0.030), control mean 0.33. Section 2.4 (p.15) gives 33.1% control and 33.7% treatment 'by the second endline survey'. Searching the text for 38%/42%/'ever borrow' found no alternative definition; the 42% on line 426 concerns businesses. So the conclusion's 38% overall and the 42% vs 33% split (a 9-point gap) do not match the reported endline-2 borrowing. Footnote 30 also has the doubled unit '33% percent'.

Second check: another model agreed

Report a problem with this finding

This finding is

wrong right unclear or overstated already corrected (erratum, later version)

I am an author of this paper Send

• p. 53 Table 10 India row col 7

  3% for number of business started (p<0.05)

• p. 24 Section 4.2

  30% more business were created in treatment than in comparison areas

• p. 47 Table 5 Panel A col 2

  0.016**

• p. 24 Section 4.2

  (30% more business were created in treatment than in comparison areas)

Judge's note: Table 10 note (2) says India effect sizes are percentages of endline control group means. Table 5 Panel A col 2 shows Treatment 0.016** and Mean in control 0.053 (p047 upright image). That is a 30% effect, matching the text on p24 (5.3 vs 6.9 per 100 households, 30% more). The printed 3% looks like a dropped digit.

Second check: another model agreed

Report a problem with this finding

This finding is

wrong right unclear or overstated already corrected (erratum, later version)

I am an author of this paper Send

• p. 44 Table 2 Panel A, Household is renter

  Household is renter 0.089 0.087 0.002

Only the authors can say what is right here.

Judge's note: In the p.44 image, nearly every other row's column 3 equals col 2 minus col 1 to within 0.001 (e.g. household size, TV, cell phone, income, loans). Rounding cannot turn 0.087-0.089 positive. Only the authors can say whether the sign or the levels are wrong.

Second check: another model agreed

Report a problem with this finding

This finding is

wrong right unclear or overstated already corrected (erratum, later version)

I am an author of this paper Send

• p. 53 Table 10, India col 3

  positive insig (endline 1 and 2)

• p. 46 Table 4 Panel B col 1

  -48.3

• p. 53

  postivie, insig, endline 1;

Judge's note: Table 4 (p46 upright) shows Panel A col 1 = 10.1 (positive) and Panel B col 1 = -48.3 (negative). Section 4.1 says endline 2 effects on total and non-durable spending are 'negative with t-statistics around 1'. The Table 10 col 4 entry (negative, both endlines) is consistent with Table 4, but col 3 is not, for endline 2. The 'postivie' spelling point duplicates C078.

Second check: another model agreed

Report a problem with this finding

This finding is

wrong right unclear or overstated already corrected (erratum, later version)

I am an author of this paper Send

• p. 5 Footnotes 2 and 3

  This approach was pioneered by Karlan and Zinman (2009)

• p. 5 Footnote 3

  case of Karlan and Zinman (2010).

• p. 38 References

  Karlan, D. and J. Zinman (2010a). Expanding credit access

• p. 38 References

  Karlan, D. and J. Zinman (2010a). Expanding credit access: Using randomized supply deci-

Judge's note: Checked footnotes 2–3 on the p5 image and the reference list on the p38 image: 2010a (RFS, South Africa), 2010b (Manila IPA working paper), 2011 (Science). There is no 2009 entry, and the in-text years never use the a/b suffixes.

Second check: another model agreed

Report a problem with this finding

This finding is

wrong right unclear or overstated already corrected (erratum, later version)

I am an author of this paper Send

• p. 11 Footnote 10

  inspired by Dinardo, Fortin and Lemieux (2010)

• p. 38 References

  DiNardo, J., N. M. Fortin, and T. Lemieux (1996).

• p. 11 footnote 10

  Dinardo, Fortin and Lemieux (2010)

• p. 38 References

  DiNardo, J., N. M. Fortin, and T. Lemieux (1996)

Judge's note: Footnote 10 on page 11 gives the year as 2010. The only matching reference-list entry (page 38) is dated 1996 (Econometrica 64(5)). The capitalisation also differs: 'Dinardo' in the footnote, 'DiNardo' in the list.

Second check: another model agreed

Report a problem with this finding

This finding is

wrong right unclear or overstated already corrected (erratum, later version)

I am an author of this paper Send

• p. 17 Section 3.2

  intersect the 45 degree line, OO′ at points E and E′

• p. 40 Figure 2

  Figure 2:

Judge's note: In the full Figure 2 image on p.40, the label at the U′ intersection with the 45° line, above E, reads E″ with a double prime. The text on p.17 calls it E′.

Second check: another model agreed

Report a problem with this finding

This finding is

wrong right unclear or overstated already corrected (erratum, later version)

I am an author of this paper Send

• p. 20 Section 3.4

  as the baseline happened roughly 15 to 18 months after Spandana started its operation in each slum

• p. 20 Section 3.4

  as the baseline happened roughly 15 to 18 months after Spandana started its operation

• p. 20 Section 3.4

  the baseline happened roughly 15 to 18 months after Spandana started

• p. 20

  as the baseline happened roughly 15 to 18 months after Spandana

Judge's note: p.20 image. Section 2.2 (p.10) says the first endline survey was conducted at least 12 months, and generally 15 to 18 months, after Spandana began disbursing loans. The baseline (2005) came before Spandana's expansion in 2006-2007. The same sentence's first half defines the cycle as baseline to endline 1.

Second check: another model agreed

Report a problem with this finding

This finding is

wrong right unclear or overstated already corrected (erratum, later version)

I am an author of this paper Send

• p. 21 Footnote 16

  Table A1 shows that treatment and comparison areas are balanced in terms of these characteristics

• p. 54 Table A1

  Treatment-Control balance in fixed characteristics (Endline 1)

Only the authors can say what is right here.

Judge's note: The controls listed on p.21 are area population, total businesses, average per capita expenditure, and the literacy of heads and of adults. Table A1 (p.54) has spouse literate, spouse works for a wage, household size, prime-aged women, teens, old businesses, land, and adult labor supply. Its title is 'Endline 1'. Table 1 is the baseline-balance table (p.9 text). Only the authors can say which table was meant.

Second check: another model agreed

Report a problem with this finding

This finding is

wrong right unclear or overstated already corrected (erratum, later version)

I am an author of this paper Send

• p. 22 Section 4.1

  about Rs. 9 per family per month (column 3)

• p. 22 Footnote 18

  8.73 is the reduction on temptation good spending per capita per month

• p. 46 Table 4 header

  Monthly (per capita)

• p. 22 Section 4.1

  reduced by about Rs. 9 per family per month (column 3)

Judge's note: Table 4 (p.46) puts col 3 (-8.73*) under 'Monthly (per capita)'. Footnote 18 multiplies 8.73 by household size 4.68 and by 12. A per-family monthly reduction would be about Rs. 41.

Second check: another model agreed

Report a problem with this finding

This finding is

wrong right unclear or overstated already corrected (erratum, later version)

I am an author of this paper Send

• p. 22 Section 4.1

  festival spending per capita in the previous year (Rs. 763

• p. 46 Table 4 header, col 5

  Yearly (total)

Judge's note: In Table 4 (p.46) Festivals (col 5, -763*) sits under 'Yearly (total)'. Footnote 18 computes 1255 = 763 + 8.73*4.68*12 per family per year, so 763 is treated as a household total.

Second check: another model agreed

Report a problem with this finding

This finding is

wrong right unclear or overstated already corrected (erratum, later version)

I am an author of this paper Send

• p. 25 Section 4.2

  are 0.015 percentage points more female-owned businesses

• p. 47 Table 5 col 3

  0.015***

• p. 25 Section 4.2

  are 0.015 percentage points more female-owned businesses in treatment than in control areas, an increase of 58%

• p. 25 Section 4.2

  0.015 percentage points more female-owned businesses

Judge's note: Table 5 col 3 (p.47) shows 0.015*** with a control mean of 0.026, and 0.015/0.026 = 58%, which matches the text's percentage. A difference of 0.015 percentage points would not give 58%. This is a unit slip.

Second check: another model agreed

Report a problem with this finding

This finding is

wrong right unclear or overstated already corrected (erratum, later version)

I am an author of this paper Send

• p. 53 Table 10 notes

  India results are from Banerjee et al. (2012).

• p. 37 References

  Banerjee, A., A. Chandrasekhar, E. Duflo, and M. Jackson (2012). The diffusion of microfi-

• p. 37 References

  Banerjee, A., A. Chandrasekhar, E. Duflo, and M. Jackson (2012). The diffusion of microfinance.

• p. 37 References

  The diffusion of microfinance. MIT working paper.

Only the authors can say what is right here.

Judge's note: The reference list (p37) has only two Banerjee entries: Banerjee (2013) in the Annual Review of Economics, and Banerjee, Chandrasekhar, Duflo and Jackson (2012) on the diffusion of microfinance. The India (BDGK) row reports this paper's own Hyderabad results (e.g. 17.46% durables, 48.3% MFI). The citation therefore points to the wrong work, and the intended source (probably this paper or an earlier version of it) is not in the reference list. Only the authors can say which reference was meant.

Second check: another model agreed

Report a problem with this finding

This finding is

wrong right unclear or overstated already corrected (erratum, later version)

I am an author of this paper Send

• p. 46 Table 4 Panel B

  0.0054

Only the authors can say what is right here.

Judge's note: Visible in the p.46 upright image: Panel A labels its coefficient row 'Treatment', but Panel B does not, and an extra row sits below the SEs. Its meaning cannot be determined from the paper.

Second check: another model agreed

Report a problem with this finding

This finding is

wrong right unclear or overstated already corrected (erratum, later version)

I am an author of this paper Send

• p. 48 Table 6 cols 4-5, control mean and stdev

  0.42 0.12 / 1.74 0.33

Only the authors can say what is right here.

Judge's note: Checked the p.48 image. A binary variable's SD is at most 0.5, and sqrt(0.12*0.88) is about 0.325, matching the 0.33 in column 5. Table 7 shows the expected pattern. The coefficients may also be affected; only the authors can say which.

Second check: another model agreed

Report a problem with this finding

This finding is

wrong right unclear or overstated already corrected (erratum, later version)

I am an author of this paper Send

• p. 53 Table 10 India row col 16

  11% (p<.05)

• p. 50 Table 8 Panel A col 3

  2.78*

Judge's note: The magnitude matches the same estimate (2.78/25.8 is about 11%). The t-ratio of about 1.88 is consistent with Table 8's single star, not p<.05. Column 15's p<0.05 matches Table 8 col 1 (3.22**). Checked both images.

Second check: another model agreed

Report a problem with this finding

This finding is

wrong right unclear or overstated already corrected (erratum, later version)

I am an author of this paper Send

• p. 6 Introduction

  The me- dian marginal new business is both less profitable and less likely to have even one employee

• p. 36 Conclusion

  we find that the average new business in a microcredit treatment area less likely to have an employee

• p. 36 Conclusion

  this decrease in not due to a few

• p. 36 Conclusion

  business in a microcredit treatment area less likely to have an employee

Judge's note: The p36 image shows all three conclusion typos. The p6 image shows the introduction sentence. Section 4.2 (main.txt l.895–900) says the median result is for profits (Table 7 col 2) and that the 'average new business' is less likely to have employees (col 5), so 'median' in the introduction misdescribes the employee result.

Second check: another model agreed

Report a problem with this finding

This finding is

wrong right unclear or overstated already corrected (erratum, later version)

I am an author of this paper Send

• p. 55 Table A2 note (1)

  Panel B presents the coefficient from regressing a dummy for "attrited between endline 1 and endline 2" on various endline

• p. 10 Section 2.2

  higher per capita expenditure at endline 1, by Rs. 131 (column 1)

Judge's note: In the p55 image, Panel B is labelled 'Attrited' with a 'Non-attriter mean' row, and the coefficients are in the units of each characteristic (Rs, shares). The text on p10 reads the coefficient as attritors having 'higher per capita expenditure at endline 1, by Rs. 131'. The note describes the regression in the reverse direction.

Second check: another model agreed

Report a problem with this finding

This finding is

wrong right unclear or overstated already corrected (erratum, later version)

I am an author of this paper Send

• p. 18 Derivative of vd

  δu(y+cd−rb)

• p. 19 Labor-supply objective

  δ[u(y + cd −rb + wl2) −v(l2)

• p. 18 Section 3.2 derivative equation

  u′(y+b)−δru(y−rb)

• p. 18 Section 3.2 derivative

  δu(y+cd−rb)]

Only the authors can say what is right here.

Judge's note: On p.18 the expression inside the d/db bracket reads δu(y+c_d−rb), while its derivative on the same line reads δru′(y+ac_d−rb). Point A on p.17 is (y−(1−a)cd, y+acd), and the appendix (p.59) uses y+acd−rb. On p.19 the labor-supply objective again has y+c_d−rb+wl2, and its bracket is not closed; the companion line closes with ']'. Reworked: the right-hand side derivative is correct for y+acd, so the factor a is dropped only in the objective. The result stands, hence S3. The missing prime and the duplicated words are judged under C034 and C035.

Second check: another model agreed

Report a problem with this finding

This finding is

wrong right unclear or overstated already corrected (erratum, later version)

I am an author of this paper Send

• p. 18 Section 3.2

  max{u′(y+b)−δru(y−rb), 0}

Only the authors can say what is right here.

Judge's note: Reworked: d/db[u(y+b)+δu(y−rb)] = u′(y+b) − δr u′(y−rb). The page image on p.18 shows u without a prime in the second term, whereas the v_d derivative on the same page has u′. This is a typographical slip in a derivation step, and the stated result stands.

Second check: another model agreed

Report a problem with this finding

This finding is

wrong right unclear or overstated already corrected (erratum, later version)

I am an author of this paper Send

• p. 36 Conclusion, para 4

  They continue to do so several years later

• p. 46 Table 4 Panel B cols 4–5

  62

• p. 23 Section 4.1

  The effect on festivals is now positive but nowhere near being significant.

• p. 46 Table 4 Panel B

  -9.99

Only the authors can say what is right here.

Judge's note: Checked the conclusion crop, the Section 4.1 crop (p.23) and Table 4 on p.46. Section 4.1 itself says the festival effect "is now positive but nowhere near being significant", that the gap "has shrunk", and that "there is no difference on average in durable goods spending in endline 2". Only the temptation-goods point estimate stays negative, and it is not significant. So 'continue to do so' overstates persistence for the three categories it refers to. This is a wording issue; the authors should qualify the sentence.

Second check: another model agreed

Report a problem with this finding

This finding is

wrong right unclear or overstated already corrected (erratum, later version)

I am an author of this paper Send

• p. 25 Section 4.2

  increase of Rs. 357 per month in business revenues

• p. 47 Table 5 col 11

  357

• p. 25 Section 4.2

  an increase of Rs. 357 per month in business revenues

• p. 47 Table 5 col 11

  Business profits

Judge's note: Table 5 (p.47) Panel A shows col 9 Business revenue = 920 and col 11 Business profits = 357. Earlier in the same paragraph on p.25, 357 is called profits.

Second check: another model agreed

Report a problem with this finding

This finding is

wrong right unclear or overstated already corrected (erratum, later version)

I am an author of this paper Send

• p. 33 Section 5

  which is not present in any of the other studies except AADFH

• p. 53 Table 10 Mongolia col 6

  negative insig

Only the authors can say what is right here.

Judge's note: Checked Table 10 continuation (upright p.53): the Mongolia col 6 entry is 'negative insig'. The later sentence on AADFH ownership of VCRs etc. concerns stocks, not purchases in the last year. Text and table disagree, and only the authors can say which is right.

Second check: another model agreed

Report a problem with this finding

This finding is

wrong right unclear or overstated already corrected (erratum, later version)

I am an author of this paper Send

• p. 33 Section 5

  In all studies except CDDP, this expansion does not translate into significantly higher profit on average.

• p. 52 Table 10 Morocco col 10

  positive, insig

Only the authors can say what is right here.

Judge's note: Table 10 col 10 (pp.52–53): Bosnia positive insig, Mexico negative insig, Morocco positive insig, Mongolia negative insig, India positive insig. No study, CDDP included, shows a significant mean-profit effect in the table, so the exception for CDDP is not supported by the paper's own summary.

Second check: another model agreed

Report a problem with this finding

This finding is

wrong right unclear or overstated already corrected (erratum, later version)

I am an author of this paper Send

• p. 59 Appendix 1, mean value theorem

  y2 ∈ (y + acd, y + acd + b)

Only the authors can say what is right here.

Judge's note: Reworked on the p59 image. The two arguments are y+acd−rb and y+acd+b, separated by b+rb, which matches the printed coefficient δ(b+rb). The lower bound of the interval is therefore y+acd−rb, not y+acd. The next step, 'y2 ≥ y1 (because b<cd)', follows from the printed interval: y+acd ≥ y+b−(1−a)cd iff cd ≥ b. With the correct interval it needs y+acd−rb ≥ y+b−(1−a)cd, i.e. cd ≥ (1+r)b, which b<cd does not imply. The step as written is faulty and needs an extra condition or argument. The stated conclusion may still hold under such a condition.

Second check: another model agreed

Report a problem with this finding

This finding is

wrong right unclear or overstated already corrected (erratum, later version)

I am an author of this paper Send