Why does female genital cutting (FGC) persist in certain places while has declined elsewhere? Using survey data from the Gambia, we study an important aspect of the persistence of FGC, namely the relationship between (i) whether a woman has undergone FGC and (ii) her support for the practice. Our data exhibit sufficient intrahousehold variation in both FGC status and in support for the practice to allow controlling for unobserved heterogeneity between households. First, our results suggest that a woman who has undergone FGC 40 percentage points more likely to be in favor of the practice, from a baseline likelihood of 40%. Second, our findings indicate that 85% of the relationship between whether a woman has undergone FGC and her support for the practice can be attributed to individual- or household-level factors, but that only 15% of that relationship can be explained by factors at the village level or beyond. This suggests that village-wide pledges against FGC, though they have worked well in neighboring Senegal, are unlikely to be effective in the Gambia. Rather, policies aimed at eliminating FGC in this context should instead target individuals and households if they are to be effective.

That’s the abstract of my most recent working paper (see here for the RepEc version, and here for the SSRN version), “All in the Family: Explaining the Persistence of Female Genital Cutting in The Gambia,” which my former Masters student Tara Steinmetz (who was a Peace Corps volunteer in The Gambia) and I have been working on for quite some time. A previous version had been circulated for the Midwest International Economic Development Conference, but this one is considerably improved. As with any working paper, the caveat that these results have not yet been through the peer-review process applies.

As the abstract above indicates, what is really nice about our data is that they exhibit enough within-household variation in FGC to allow holding household-level factors constant for the women within a given household. This then allows quantifying the contribution of each level of variation — individual, household, village, and district — to the persistence of FGC, defined here as the relationship between a woman having undergone FGC and whether she supports the practice.

Our findings suggest that in The Gambia:

1. The average woman aged 15-49 is 40% likely to support FGC. Having experienced FGC, however, makes a woman twice as likely to support FGC, i.e., it increases the likelihood that she will be in favor of FGC to about 80%.
2. About 85% of FGC persistence can be explained by individual- and household-level factors, and only 15% of it can be explained by factors at or beyond the village level (e.g., district, region, etc.) The following figure illustrates that point.

(Source: Bellemare and Steinmetz, 2013.)

This does not bode well for some of the policies that have so far been used to eliminate FGC in West Africa. In Senegal, which encompasses almost all of The Gambia, village-wide pledges to stop practicing FGC have seemingly been effective in phasing out the practice; similar village-wide pledges had previously been effective against foot-binding in China (see Mackie’s fascinating 1996 American Sociological Review article). Given our results, however, there is little hope that similar village-level pledges will work to phase out FGC in The Gambia: if 85% of the action occurs at the individual and household levels, targeting the village level does not seem like the most promising idea.

[T]he events of recent years have brought into stark relief the great challenges that society faces and the role for agricultural economists in helping to meet them. The agricultural productivity growth that enabled food supply to grow faster than demand—and on a shrinking land base—has slowed, contributing to the recent rises in commodity prices. Changes in climate will present further challenges to sustaining productivity growth, but public R&D investments are languishing in many places. World population may increase by one-third by 2050, and rapid economic growth in China and India, home to more than one-third of the world’s population, has caused dramatic changes in diets and food demands in those countries. Along with changes in food demand, new demands for biofuels are now competing for grain. In short, agriculture is challenged to meet rapidly growing demands for food, feed, and fuel, and to do so with ever-smaller environmental impact.

Food demands are not only growing, they are changing in ways most of us would not have imagined. The attributes that define food products and production practices have expanded rapidly. In addition to traditional product attributes such as taste, appearance, convenience, brand appeal, and nutrition, consumers increasingly care also about aspects of the production process (e.g., use of chemicals, farm location and size, and treatment of animals), marketing arrangements (in particular, their “fairness”), and implications of food production and consumption for the environment.

Indeed, within this macro environment confronting agriculture lie countless puzzles, contradictions, and fascinating and important research questions that demand answers only we can provide.

Wise words from Rich Sexton, president of the Agricultural and Applied Economics Association (AAEA), in his column for The Exchange, the AAEA’s newsletter.

(Credit: xkcd.)

Last Monday morning, when my wife left for work, I told her I was expecting a meager number of page views that day given my choice of post topic. I was wrong: people really care about binary dependent variables.

The post generated quite a bit of commentary. Some said that if you have experimental data, you would not want to run a regression. But that’s not completely true. Sure, with experimental data, you can run a t-test comparing the means of the control and treatment groups. But I can think of many cases where you would still want to run a regression in order to increase the precision of your estimate of the treatment effect.

The most interesting response came from Penn State’s Christopher Zorn in a post on his blog. If you want to read the remainder of this post, I suggest you read his post and come back for mine (and make sure to add his blog to your RSS feed while you’re at it).

Done? Okay, here goes:

1.  I did not read the King and Roberts working paper Christopher links to (with my upcoming move halfway across the country, I need all the time I can get to work on my own research), but in the comments to my post, Conner responds “The King and Roberts results are more relevant for the case when identification of all parameters of interest requires that we have the correct model, e.g., forecasting probabilities. This isn’t the case when looking at binary treatment assignment and are interested in estimating average treatment effects. You just need the expectation of the error term to be the same in the treatment and control groups. King and Roberts more or less make this point themselves on page 3 of their paper.” Moreover, the probit standard errors model one kind of variance (that due to the Bernoulli structure of the dependent variable), but they are not robust to other kinds of heteroskedasticity. And with heteroskedasticity, the probit and logit coefficients are inconsistent, even with robust standard errors (ht: Tim Beatty.)
   
2. I must insist that forecasting probabilities is not what I am interested in. Most of the time, I’m interested in getting as close as possible to the average treatment effect. If you are interested in forecasting probabilities, then by all means estimate a probit or a logit.
3. On nonlinear functional forms, in the example Christopher gives (the likelihood that extremely poor or extremely wealthy people will purchase a TV will not change much if their income increases by $1,000, but the change will be much larger for someone with an average income), the nonlinear function is best modeled by including both income and its square, so as to model nonlinearities in the impact of income on the likelihood that someone purchases a television. But even then, this assumes that we know the exact shape of the nonlinear relationship.

And in response to Christopher’s last two points, I do not use R, but I plugged his example in Stata. What happens here is that the LPM will give a coefficient estimate of 0.6, but the logit omits x altogether because in those cases where x = 1, y is perfectly predicted, i.e., whenever x = 1, y = 1. Having compaed the LPM with fixed effects with the conditional logit with fixed effects for one of the applications I have worked on, the latter does the same thing.

I don’t know that not dropping those observations is a bad thing, though: Even in Christopher’s example, those observations contain a lot of information about the relationship between x and y, namely that the two are highly positively correlated. To see this, supposed you wanted to know the likelihood that people who smoke will die of lung cancer. You collect individual data on smoking and on causes of death, and you find that everyone who smokes dies of lung cancer, but that only about half of the people who don’t smoke die of lung cancer.

If you wanted to know how the decision to start smoking changes the likelihood that someone will die of lung cancer, would you throw away all the observations for which an individual is a smoker? I wouldn’t, as they contain valuable information that help us quantify the marginal impact of the decision to smoke on the likelihood of dying from lung cancer.

With that said, I add the caveat that I am not an econometrician and that, to paraphrase a soon-to-be-colleague, I have strong opinions that are weakly held. It looks as though one’s preferred estimator for binary dependent variables is really all a matter of disciplinary cultural norms (economist love probits; other social scientists, not so much), if not of field cultural norms within disciplines.

So ultimately (and this deserved to be in bold), because no estimator is perfect, you should you always estimate all three (LPM, probit, and logit) and compare their results to make sure nothing is amiss.

I would just add:
(1) Granger-causality is not causality as is typically understood, and is best interpreted as “predictability.” Thus, if knowledge of speculative positions does not improve our price forecasts, it is reasonable to conclude that speculation does not affect prices.
(2) Speculators can profit either by taking long (buy) or short (sell) positions, so there is no reason to believe they would have an interest in strictly rising prices.
(3) Prices have gone up for some commodities for which there are no financial instruments (i.e. speculation); and prices have not gone up for some commodities for which there are financial instruments.

The “Masters Hypothesis” is the claim that unprecedented buying pressure from new financial index investors created a massive bubble in agricultural futures prices at various times in recent years. This paper analyzes the market impact of financial index investment in agricultural futures markets using non-public data from the Large Trader Reporting System (LTRS) maintained by the U.S. Commodity Futures Trading Commission (CFTC). The LTRS data are superior to publicly-available data because commodity index trader (CIT) positions are available on a daily basis, positions are disaggregated by contract maturity, and positions before 2006 can be reliably estimated. Bivariate Granger causality tests use CIT positions in terms of both the change in aggregate new net flows into index investments and the rolling of existing index positions from one contract to another. The null hypothesis of no impact of aggregate CIT positions on daily returns is rejected in only 3 of the 12 markets. Point estimates of the cumulative impact of a one standard deviation increase in CIT positions on daily returns are negative and very small, averaging only about two basis points. The null hypothesis that CIT positions do not impact daily returns in a data-defined roll period is rejected in 5 of the 12 markets and estimated cumulative impacts are negative in all 12 markets; the opposite of the expected outcome if CIT rolling activity simultaneously pressures nearby prices downward and first deferred prices upward. Overall, the results add to the growing body of literature showing that buying pressure from financial index investment in recent years did not cause massive bubbles in agricultural futures prices.

From a new NBER working paper (ungated version here) by Nicole Aulerich, Scott Irwin, and Philip Garcia. Here is a discussion of the paper by Aaron Smith, who writes:

I commend [Aulerich et al.] for emphatically refuting the Masters Hypothesis. It is important for economists to bring facts and rigorous data analysis to bear on issues that have received such publicity, especially if they appear to be influencing policy

In short, the Aulerich et al. findings fly in the face of what many food policy “experts” love to believe about food prices.

I will dedicate a post to that paper when we have a manuscript that is presentable, but I wanted to talk about the “developmental paradox,” since this is something that has been coming up frequently in my research and teaching, and because most readers of this blog are probably unaware of the paradox.

In Food Politics, Paarlberg (2011) writes:

The governments of nearly all rich countries provide subsidies to support the income of farmers. In 2006, according to calculations by the Organization for Economic Co-operation and Development (OECD), government policies in these rich countries transferred $283 billion worth of income to farmers … Roughly 29 percent of all farm earnings in these countries depended on such government programs. …

Governments in poor developing countries provide much less subsidy support to agriculture despite being far more “agricultural.” In fact, poor countries often make a practice of taxing their farmers to help finance subsidies for urban food consumers. They rig their internal markets to oblige farmers to sell food at an artificially low price, thus creating an income transfer away from farmers and toward food consumers. So while policies in rich countries tend to be rural biased, policies in poor countries tend to be urban biased.

That is the essence of the developmental paradox: the fact that as countries develop, agricultural protection increases even though the share of agriculture usually declines along the path to development.

The relationship holds both in time series and in cross-sectional data, and it was first identified by Lindert (1991), but Chris Barrett has a very nice 1999 article looking at the microeconomics of the developmental paradox.

There are many explanations for the developmental paradox, all of them have probably been true at some point. In developing countries, governments tax producers and subsidize consumers to reward urban elites and prevent social unrest. Indeed, the Arab Spring began with food riots in Algeria, in response to which the Algerian government announced more food subsidies. At the first few demonstrations in Tunisia, people were brandishing loaves of bread. And many believe that the Mubarak regime fell when the Egyptian government could no longer subsidize bread.

In developed countries, it gets a bit more complicated. The geography of political representation in some countries gives rural districts too much power relative to their size (think of how Iowa and Nebraska both have two senators, just like California and New York), and thus leads to farmers’ interests being over-represented. Or the smaller the size of the agricultural sector, the easier it is for farmers to organize into farmer organizations and lobby the government. Paarlberg cites evidence that the smaller commodity groups (e.g., sugar producers) are the ones getting the most out of agricultural protection.

If you were interested in knowing what the effect of attending the information session on the likelihood that someone is a smoker, i.e., the impact of on  The equation of interest in this case is

(1) ,

where  is a constant,  is a vector of the coefficients attached to the explanatory variables,  is the parameter of interest, and  is the error term.

This post is about why, in most cases, you should be estimating equation (1) by ordinary least squares, i.e., estimate a linear probability model (LPM). I have heard and read so many arguments against the LPM and for the probit or logit that I wanted to write something on this.

Arguments Typically Made against the LPM

The arguments typically made against the LPM are:

1. The error term of a binary variable has a Bernoulli structure, i.e., , where . This non-constant variance of the error term means that you have a heteroskedasticity problem and the LPM standard errors will be wrong.
2. The LPM can yield values of , i.e., predicted values of the dependent variable, outside of the  interval. In other words, the LPM can yield predicted probabilities that are negative or greater than 100%.
3. The LPM imposes linearity on the relationship between the dependent variable and the right-hand side variables.

For these reasons, many will dismiss LPM estimates as wrong. I respond:

1. With robust standard errors, the standard errors are correct, and it is very easy it is to implement robust standard errors with most statistical packages. Indeed, in the package that I use the most, it is simply a matter of adding “, robust” at the end of my estimation command.
2. This is only a concern if your reason for estimating equation (1) is to forecast probabilities. For most readers of this blog, that will not be why they are estimating equation (1). Rather, they will be interested in knowing the precise value of .
3. Sure, but who says an assumed nonlinear relationship is much better? On their Mostly Harmless Econometrics blog, Angrist and Pischke write:
   

   If the conditional expectation function (CEF) is linear, as it is for a saturated model, regression gives the CEF – even for LPM. If the CEF is non-linear, regression approximates the CEF. Usually it does it pretty well. Obviously, the LPM won’t give the true marginal effects from the right nonlinear model. But then, the same is true for the “wrong” nonlinear model! The fact that we have a probit, a logit, and the LPM is just a statement to the fact that we don’t know what the “right” model is. Hence, there is a lot to be said for sticking to a linear regression function as compared to a fairly arbitrary choice of a non-linear one! Nonlinearity per se is a red herring.

Arguments One Can Make Against the Probit or Logit

People who dismiss the LPM, usually by invoking the two arguments above, usually argue in favor of estimating a probit or a logit instead. Here are some arguments one can make against the probit or logit:

1. Both the probit and the logit can lead to identification by functional form. If you are interested in identifying the causal relationship flowing from to , i.e., in precisely estimating , you want to avoid this.
2. The probit and logit are not well-suited to the use of fixed effects because of the incidental parameters problem.

Quo Vadis?

So when should you use an LPM, and when should you use a probit or a logit? If you have experimental data, i.e., if values of  were randomly assigned, there is no harm in estimating a probit or a logit — your estimate of  is cleanly identified because of the random assignment. If you want to forecast the likelihood that something will happen, estimate a probit or a logit.

But if you are interested in estimating the causal impact of on  and have any reason to believe that your identification is less than clean, if you want to use fixed effects, and if you are not interested in forecasting the value of , you should prefer the LPM with robust standard errors.

Conclusion

I have made the points above several times over the last few years, in conversations with colleagues, when advising students, in referee reports, etc. But every once in a while, I will get admonished by an anonymous reviewer for my use of the LPM, and so I wanted to write something about it.

Ultimately, I think the preference for one or the other is largely generational, with people who went to graduate school prior to the Credibility Revolution preferring the probit or logit to the LPM, and with people who went to graduate school during or after the Credibility Revolution preferring the LPM.

As always, the right way to approach things is probably to estimate all three if possible, to present your preferred specification, and to explain in a footnote (or show in an appendix) that your results are robust to the choice of estimator.

Urban agriculture may have an important role to play in addressing food insecurity problems, which are bound to become increasingly vital with the secular trends towards the urbanization of poverty and of population in developing countries. Our understanding of the importance, and food security implications of urban agriculture is however plagued by a lack of high quality, reliable data. While studies based on survey research data do exists for several major cities, much of the evidence is still qualitative if not anecdotal. Using a recently created data set bringing together comparable, nationally representative household survey data for 15 developing or transition countries, this paper analyzes in a comparative international perspective the importance of urban agriculture for the urban poor and food insecure. On the one hand, the potential for urban agriculture to play a substantial role in urban poverty and food insecurity reduction should not be overemphasized, as its share in income and overall agricultural production is often quite limited. On the other hand, though, its role should also not be too easily dismissed, particularly in much of Africa agriculture provides a substantial share of income for the urban poor, and for those groups of households for whom it constitutes an important source of livelihood. We also find fairly consistent evidence of a statistical association between engagement in urban agriculture and dietary adequacy indicators.

In other words, though those who engage in urban agriculture are better off from it, urban agriculture itself does not look like it can ensure food security. From a new working paper by the University of Rome Tor Vergata’s Sylvester Jatta.

At the industrial park, female workers wearing chartreuse aprons and headscarves stream out of the blue factory buildings on their lunch break. Frandline Joseph sits outside. She sews for Sae-A and says she doesn’t like the work: “I don’t have time to sit.”

But she also says that she had no job before her current one, and life has improved since finding employment. “Now I work for 200 gourdes,” [Note: $5 daily -- MFB.] she says, and can pay her daughter’s school fees in a country with a virtually non-existent public education system. “Before the park, I worked for nothing.”

Her story is similar to other published accounts, and that of Rosedaline Jean, a 22-year-old who’s worked for Sae-A for five months. “Before, I lived only by the grace of God,” says Jean. “Although I don’t have a husband or children, my life wasn’t easy because I wasn’t working. When I got here, a lot changed in my life.

“This isn’t the ideal job,” she continues, “but it’s better than nothing. I don’t intend to make a career in this job. I plan to start a business, and I’m already saving for it. But it’s difficult, because my salary is practically nothing.”

From an article by Tate Watkins in The Atlantic.

I am often suspicious of industrial policy, but in this case, it looks as though manufacturing might be the key to Haiti’s economic development. Indeed, the country has some agriculture, but it does not see to have the infrastructure necessary to develop the agricultural value chains necessary to rely on agricultural exports as a driver of development.

Because value chains are considerably shorter in manufacturing and are thus are less likely to fail because of a weak link (as agricultural value chains are wont to do), and because the transportation costs between the US and Haiti are very low, there are good reasons to be optimistic. Still, the industrial park could turn out to be a very costly development policy experiment.

But if Haiti truly wants to rely on manufacturing as an engine of development, it will also have to invest in public education, as there is only so much an uneducated workforce can make.

As a side note, Tate Watkins is perhaps my favorite among the handful of journalists who write on development-related issues. He is also one of the few journalists I know of who have a solid understanding of economics.

This paper reports on the first randomized evaluation of the impact of introducing the standard microcredit group-based lending product in a new market. In 2005, half of 104 slums in Hyderabad, India were randomly selected for opening of a branch of a particular microfinance institution (Spandana) while the remainder were not, although other MFIs were free to enter those slums. Fifteen to 18 months after Spandana began lending in treated areas, households were 8.8 percentage points more likely to have a microcredit loan. They were no more likely to start any new business, although they were more likely to start several at once, and they invested more in their existing businesses. There was no effect on average monthly expenditure per capita. Expenditure on durable goods increased in treated areas, while expenditures on “temptation goods” declined. Three to four years after the initial expansion (after many of the control slums had started getting credit from Spandana and other MFIs), the probability of borrowing from an MFI in treatment and comparison slums was the same, but on average households in treatment slums had been borrowing for longer and in larger amounts. Consumption was still no different in treatment areas, and the average business was still no more profitable, although we find an increase in profits at the top end. We found no changes in any of the development outcomes that are often believed to be affected by microfinance, including health, education, and women’s empowerment. The results of this study are largely consistent with those of four other evaluations of similar programs in different contexts.

A new working paper (older, ungated copy here) by Duflo et al. The emphasis is mine.

This is consistent with another careful study (link opens a .pdf file) by Crépon et al. of the impact of microfinance in Morocco, where there authors also find that microfinance has no discernible impact on the usual development indicators (i.e., consumption, health, education, etc.)

To be sure, microfinance does appear to have some impacts, as the abstract above indicates — just not the miraculous impacts that are often touted by microfinance advocates.