Chris Woodruff and I have a recent working paper in which we look at the importance of business practices for small firms in developing countries, using surveys administered to over 20,000 small firms in Bangladesh, Chile, Ghana, Kenya, Mexico, Nigeria, and Sri Lanka. A recent Vox note summarizes some of the key results: namely (i) that better business practices are correlated with higher productivity, higher profits and higher sales in the cross-section; (ii) that firms with good business practices today are more likely to survive and grow using panel data; and (iii) combining these results with recent randomized experiments on business training suggests that these impacts are causal, and that the lack of effect of most business training programs is because they do not change business practices enough, rather than because these practices don’t matter.
I’ve received several questions about this work, and know many of our readers are interested in measurement issues, so thought I’d note some of these measurement issues in a blog post here.
I don’t want to read anything else, tell me quickly what are you measuring?
We have 26 questions that measure business practices in the areas of marketing (e.g. Did the firm advertise? Did it attempt to attract customers with a special offer? Does it ask customers what other products they would like it to sell?), record keeping (e.g. Does the firm record every sale and purchase? Has it worked out the cost of each item it sells? Does it have a written budget?), financial planning (e.g. Does it have a sales target for the next year? Does it have a balance sheet and profit and loss statement?), and buying and stock control (e.g. Does it frequently run out of stock? Does it attempt to negotiate discounts with suppliers?).
These are intended to be universal best practices, in the sense that most firms should benefit from using them. They are closely based on the goals of business training programs like the ILO’s Improve Your Business (IYB) program. They are clearly not the only practices that matter, but appear to be a useful subset.
Where can I get the questionnaire if I want to implement this in my own survey?
The questions are intended to be ones that can be asked in surveys administered to large numbers of firms, including in circumstances where phone surveys are carried out. You might then worry that firms will over-report the extent to which they are doing all these, and therefore that the self-reported data will be unreliable. We have several bits of evidence that suggest that you might not need to worry so much:
I don’t have room for 26 questions in my survey, is there a subset that are useful?
We find that 25 out of the 26 practices measured have a statistically significant correlation with labor productivity when examined one at a time (the exception being not running out of stock frequently). But if we consider the four sub-indices, we find that marketing and record-keeping have the strongest associations with profits conditional on input use, owner, and firm characteristics. The total business practice score has a correlation of 0.75 with a sub-index just covering the 7 marketing questions, and a correlation of 0.80 with a sub-index just covering the 8 record-keeping questions, and a correlation of 0.949 with an index of the 15 questions covering marketing and record-keeping. So if you are short of survey space, you could probably get away with just asking the record-keeping and marketing. A caveat here is that this will depend on the reason for measuring these in the first place: if you have an intervention that aims to change other domains of practices, you would want to measure these other domains too.
[update in response to an additional question] Do these practices still have predictive power in firms of 10 to 50 workers?
We have 903 firms in our sample with workers in this range. If I run the equivalent regression to column 1 of Table 2 in the paper (log sales on labor controls, country dummies, and country*industry dummies), restricting just to firms with 10-50 workers, the coefficient is 1.81 (p=0.000). If I run the equivalent of column 3 in the same table, the coefficient is 1.24 (p=0.008). So the overall index of practices continues to have strong predictive power (and, if anything, a larger effect size) for these somewhat larger firms.
Then if we want to know which practices seem to matter most for such firms, the record-keeping and financial planning subindices are significant when I put all four subindices together. When I look at the practices one by one, the ones with the strongest correlations with labor productivity are M6 (making special offers to customers), R1 (record-keeping), R2(recording every transaction), R7(having a written budget), R8(having records you could use for a bank loan), and F1 (review the financial performance and analyze areas for improvement at least monthly).
I’ve received several questions about this work, and know many of our readers are interested in measurement issues, so thought I’d note some of these measurement issues in a blog post here.
I don’t want to read anything else, tell me quickly what are you measuring?
We have 26 questions that measure business practices in the areas of marketing (e.g. Did the firm advertise? Did it attempt to attract customers with a special offer? Does it ask customers what other products they would like it to sell?), record keeping (e.g. Does the firm record every sale and purchase? Has it worked out the cost of each item it sells? Does it have a written budget?), financial planning (e.g. Does it have a sales target for the next year? Does it have a balance sheet and profit and loss statement?), and buying and stock control (e.g. Does it frequently run out of stock? Does it attempt to negotiate discounts with suppliers?).
These are intended to be universal best practices, in the sense that most firms should benefit from using them. They are closely based on the goals of business training programs like the ILO’s Improve Your Business (IYB) program. They are clearly not the only practices that matter, but appear to be a useful subset.
Where can I get the questionnaire if I want to implement this in my own survey?
- Chris has the business practices as used in Sri Lanka on his website
- The survey instruments and data section on my website links to different datasets I have deposited in the World Bank open data library. You can find the Bangladesh, Kenya, Nigeria, and Sri Lankan women questionnaires and full datasets there.
- A couple of the datasets we use were collected by other researchers and are not yet publicly available, but we will put a master dataset of the constructed business practice data out once the paper is published.
The questions are intended to be ones that can be asked in surveys administered to large numbers of firms, including in circumstances where phone surveys are carried out. You might then worry that firms will over-report the extent to which they are doing all these, and therefore that the self-reported data will be unreliable. We have several bits of evidence that suggest that you might not need to worry so much:
- Firms don’t all report doing everything – there is large dispersion in the business practices reported by firms, and as noted above, this dispersion does meaningfully predict business performance.
- In Sri Lanka and Nigeria we did an audit exercise, in which businesses had a 2-3 hour visit by a business mentor, who asked a lot of open-ended questions about the firm and then provided an independent assessment of whether or not the firm owner was using each of the 26 practices. The correlation between these audit reports and survey responses was 0.74. This suggests the survey reports do provide a common signal.
- You might also worry that people who have gone through business training programs might systematically over-report the use of these practices. However, we find very small treatment impacts of some business training programs on the reported use of these practices. Second, there was no difference in the deviation of the audited score from the survey score for firms who had gone through business training versus those which had not (although the audit was taken a couple of years after training).
I don’t have room for 26 questions in my survey, is there a subset that are useful?
We find that 25 out of the 26 practices measured have a statistically significant correlation with labor productivity when examined one at a time (the exception being not running out of stock frequently). But if we consider the four sub-indices, we find that marketing and record-keeping have the strongest associations with profits conditional on input use, owner, and firm characteristics. The total business practice score has a correlation of 0.75 with a sub-index just covering the 7 marketing questions, and a correlation of 0.80 with a sub-index just covering the 8 record-keeping questions, and a correlation of 0.949 with an index of the 15 questions covering marketing and record-keeping. So if you are short of survey space, you could probably get away with just asking the record-keeping and marketing. A caveat here is that this will depend on the reason for measuring these in the first place: if you have an intervention that aims to change other domains of practices, you would want to measure these other domains too.
[update in response to an additional question] Do these practices still have predictive power in firms of 10 to 50 workers?
We have 903 firms in our sample with workers in this range. If I run the equivalent regression to column 1 of Table 2 in the paper (log sales on labor controls, country dummies, and country*industry dummies), restricting just to firms with 10-50 workers, the coefficient is 1.81 (p=0.000). If I run the equivalent of column 3 in the same table, the coefficient is 1.24 (p=0.008). So the overall index of practices continues to have strong predictive power (and, if anything, a larger effect size) for these somewhat larger firms.
Then if we want to know which practices seem to matter most for such firms, the record-keeping and financial planning subindices are significant when I put all four subindices together. When I look at the practices one by one, the ones with the strongest correlations with labor productivity are M6 (making special offers to customers), R1 (record-keeping), R2(recording every transaction), R7(having a written budget), R8(having records you could use for a bank loan), and F1 (review the financial performance and analyze areas for improvement at least monthly).
Join the Conversation