Syndicate content

Measuring Entrepreneurship (II)

Markus Goldstein's picture

This post is coauthored with Francisco Campos

If it takes 5 machines 5 minutes to make 5 widgets, how long would it take 100 machines to make 100 widgets? Yes,  another question to measure entrepreneurship. Last week we talked about why we might want to measure entrepreneurship and some broad ideas about how to do it. This week we get a bit more into the details and talk about some specific ways to do it.   Keep in mind: no one has gotten rich doing this (yet), which would be the true test of these measures…

When seeking to identify entrepreneurs, the characteristics that people look for can be broken down into four main groups: tests of ability, family/personal background, measures of personality traits, and risk and time preferences.

For entrepreneurship, cognitive ability is likely to matter.   De Mel, Mckenzie and Woodruff try to capture this in Sri Lanka by using a forward digit span recall test to measure short-term cognitive processing power. The way it works is that respondents are shown a card with a four digit number. That card is taken away and ten seconds later, respondents are asked to repeat the number as written on the card. Those responding correctly are shown a five digit number, and so forth up to 11 digits. The median firm owner could recall 6 digits in this study.

De Mel and co. also use a Raven Test to measure abstract logical thinking. Respondents are provided with 12 pages, each with a 3 by 3 pattern with one cell missing. Below the pattern are eight figures, one of which fits the pattern. The patterns become progressively more difficult from the first to the 12th page. Respondents are given five minutes to complete as many patterns as possible. They find the Raven test to be more effective in predicting innovation that the digit span recall.

Another way to measure ability is to ask questions which get at cognitive reflection. De Mel and co. borrow from Frederick (ungated version here) who lays out a three question test to get at this (if you want to test yourself, don’t read Frederick’s article until you do the test on page 3).   They adapt this to the local context – since Fredrick’s test is clearly geared towards a US audience.

So while those are some options on ability, one might also want to ask some more standard survey questions that may help predict entrepreneurship. Family background (the home of many instrumental variables) is one place to look – especially parents’ education and occupational choice. Family wealth measures might also be useful. One could also use the entrepreneurs’ labor and business history to try and get at experiences that might have shaped them.

Another realm worth taking a look is personality.   This is often measured using agree-disagree scales for a series of statements. Examples of these include “Get chores done right away” and “Need a push to get started” to measure positively and negatively self-discipline. In the paper that we talked about last week, De Mel, McKenzie and Woodruff report some measures of personality.   First among these is work centrality which is captured by a question on how important work is in life.   They also capture achievement, which they measure through questions on satisfaction from doing well and a feeling of competition with others. Third, they measure power motivation, which comes from a set of questions about planning and deciding what other people should do/control over events.   The final measure is the internal locus of control which they capture through measures of willingness to take risk and to put oneself in unfamiliar circumstances.

A final realm worth exploring is attitudes towards risk and time preferences. Here a fair number of studies have used games, with real payoffs. For example Cole, Sampson and Zia, use Binswanger lotteries with real money to assess risk-aversion in India and Indonesia. Households in India were classified risk-averse if they opt to receive Rs. 2 for certain, rather than playing a lottery that paid Rs. 5. Ashraf, Karlan and Yin use standard preference games (between now and in one-month compared to 6 and 7 months from now for instance) to identify hyperbolic discounters (impatient now, patient later) and impatient individuals that work with a local bank in the Philippines. And de Mel and co. do lotteries with a 10 sided dice (Dungeons and Dragons anyone?) to measure the coefficient of relative risk aversion.  

So this is a start on some of the methods that people use.   For a review of a lot of these and what they might tell us, a site to check out is the Entrepreneurial Finance Lab (EFL) led by Asim Khwaja and Bailey Klinger. Their knowledge center has a good collection of relevant articles. In addition, they have their own proprietary psychometric test to analyze the entrepreneurs’ Ethics and Character, Intelligence, Attitudes and Beliefs, and Business Skills. This is being used by a number of banks in lieu of credit history – so maybe someone has found a way to measure entrepreneurship that you can put money behind. On personality specifically, another site worth checking out is the International Personality Item Pool.  

One set of results that can help us think about the efficacy of these measures is De Mel, McKenzie and Woodruff’s study of Sri Lankan businesses.   The fundamental idea is neat – they take a page out of biology and look at own account workers and see if these folks look more like SME owners or more like wage workers (i.e. which species are these people?).   This would, for instance, help policymakers figure out which enterprises should be targeted for assistance in growing and which ones should be helped into the wage labor pool.

 What they do is administer a set of these measures to a set of wage workers, a set of enterprise owners (where the size ranges from 5-50 employees), and then compare the answers with their set of own account workers/microenterprise owners (who have been the subject of other papers).   Their description of how the analysis works bears repeating:

“Discriminant analysis is a tool commonly used by biologists to separate animals or plants into species on the basis of easily measure characteristics.   The basic idea is to find the particular combination of the set of measured variables which best separates individuals into their distinct species. There are two main uses for this in practice. The first is similar to our logit regressions [above], studying the set of variables to characterize the nature of differences between species. The second use is to then use the fitted combination of measured variables to predict the species of new animals or plants for which only this vector of measured variables has been observed.   This is particularly useful in cases where the species of an animal can only be truly verified after exhaustive and expensive testing, possibly resulting in killing the animal.   Observing certain characteristics of the animal instead allows it to be accurately classified without taking such expensive and extreme measures. We apply these techniques to classifying another elusive animal, the own-account worker…”

And what do they find? Bottom line – this approach suggests that ability and attitudes different own account workers from SME owners better than family background indicators do.   The majority of own account workers end up looking more like wage workers than SME owners but there is a significant minority (about 1/3) who line up with SME owners – so there are some for whom business growth would very likely be a good thing. 

They also take a look at what predicts growth among the own account/microenterprise group (for whom they have a panel).   Parental background or childhood conditions are not significant.    Ability and motivation, on the other hand, seem to matter.   Folks with more ability, those motivated by personal achievement, and those willing to give up some control over their situations are more likely to grow.   And to come back to the species classification, 13.8 percent of those classified as SMEs added at least one paid employee over a 2-3 year period as compared with 7.8 percent of those classified as wage workers.   And these growth rates, they note, look good when compared to similar studies in the US. 

So these tools may be helpful in separating the higher potential businesses from those with less potential. Indeed, a recent paper by Miriam Bruhn also uses this approach to look at the effect of business registration reform in Mexico. She finds that informal business owners who classify as the formal business species are more likely to register and continue business – while those who classify as the wage worker species are more likely to become…wage workers.