How big were Haiti's farms?
More results from new data
These results are from a book chapter I’m preparing on the role of land and property rights in Haiti’s economic development. I’ll release the working manuscript in the near future, but I wanted to highlight some results before its release. Previous results are here.
The scale of farms is important for agricultural productivity. If farms are too small, it’s hard for farmers to experiment. It’s easier for one farmer with 1000 acres to internalize experiments than it is for 1000 farmers with one acre each. Smaller farms limit the scale of the agricultural technologies you can adopt. There’s no point in getting a large tractor if parking it if your farm doesn’t have enough room for it to turn. Also, small farms face lots of risk. Losing half an acre out of 1000 is nothing. Losing half an acre out of one could be devastating. Finally, small farms can’t internalize the benefits of erosion protection, so they tend to be overworked and environmental unstable.
Haiti has a history of small farms. But so has every other Caribbean country. The key difference for Haiti is that it has much lower land inequality. While other Caribbean countries had large productive farms (and in some cases, the small farms could piggyback off their scale), small farms characterize the Haitian agricultural experience.
So how large were Haitian farms?
What we know from the Census
Generally, there is very little data on Haitian land. And it seems intentional.
There’s a famous story given in Schmidt’s account of the U.S. occupation of Haiti:
A related and much more substantial impediment to plantation development was the long-standing division of land into minuscule plots held by peasant freeholders. American companies experienced great difficulty in buying up large plots of land and in gaining clear titles. The United States tried to solve this problem by court reform and by instituting a massive cadastral survey which would have permitted land consolidation since few peasants had legal documents to prove their ownership, which dated back to the division of former French colonial plantations in the early nineteenth century. Marines took aerial photographs of a large part of the country in 1925 and 1926 in preparation for the contemplated cadastral survey, but all negatives were destroyed when the building in which they were stored burned down at night in an unexplained fire.
Our understanding of Haitian farms in the mid-20th century comes from the 1950 agricultural census. As I mentioned in a previous post, the census has some problems, including possible underreporting due to fears of taxation. But it summarizes our conventional wisdom.
According to the 1950 census, the median plot in Haiti was between 1.29 and 2.58 hectares. If we split the difference, that puts it at 1.93, though I imagine most of those are right at 1.29 since that is equal to 1 carreaux, the standard unit of land measurement in Haiti.
That’s not too crazy. The median plot in the Dominican Republic in 1950 was 2.32 ha, and in Puerto Rico in 1935 it was 2.11. The big difference is that these countries had large-scale plantations while Haiti had none. But smallholder agriculture looks like it could have been pretty similar.
At least, that was what we thought. We have new data to investigate further.
New data on Haitian land size
Over the last decade, I have found new sources on land in Haiti. In my previous post, I talked about the irrigation schedules. I have also found data on a state rental program that leased state-owned land to farmers. That data has been used in my papers on the disappearance of sugar in Haiti, the investments in state capacity after a fiscal crisis during World War II, and the persistence of refugee camps on Haiti’s border. There’s data on a homesteading program, that granted state land to farmers who invested in it. The homesteading data was used in my paper on why it failed, which I also explored in a previous post. Finally, I have data on a cadaster in Artibonite. This data is part of a working paper looking at transaction costs in the land market.
That’s four new sources of data!
But when wanting to understand the size of Haitia farms, each one has a bias. Let me quickly explain them.
Irrigation schedules - since this land is irrigated, it’s more productive than the typical plot, and therefore it’s probably larger
State rental program - this land has never been farmed before or divided over generations, so we’re probably getting larger plots here
Homestead program - the program limited these plots to be at most 5 ha, but it also looks like they were targeting this size, so, again, larger
Artibonite cadaster - The cadaster only surveyed plots below 3 ha, which is above the census median, so that’s probably not a problem. But it also is in Artibonite, Haiti’s breadbasket, so it could be more productive plots, which would be larger
In all four sources of data, we have reason to believe that the farms will be larger than the average Haitian farm. That is the problem with data. As a corollary to Demsetz’s theory on property rights, we create records when the benefit of creating them exceeds the costs, which means we’re more likely to get records on larger farms.
Nevertheless, we persist. Let’s stack up the data.
Results on Haitian farms
In the figure below, I plot the distribution of farm sizes in each data set. The census is the top one, and the dashed line marks where the census median is.
This figure surprised me.
Even though there’s reason to believe that each of the new data sets come from positively-selected plots, the median for three out of the four distributions is below the census median. For the irrigation schedules and land rentals, the median is 1.29 ha. That’s actually probably pretty close to the census median since in the census it’s somewhere between 1.29 and 2.58. But the Artibonite cadaster is a really small 0.50 ha. That’s only about 25-40% of the census median.
One problem with the Artibonite cadaster is that it only surveyed land smaller than 3 ha. Maybe that land above 3 ha would pull the median up. But the benefit of the Artibonite cadaster is that we know exactly what the selection is. The other data sets have hypothesized biases. But we can apply the same 3 ha cutoff to the census data and do a direct comparison to see how close the distributions match.
Below I compare the Artibonite distribution to the census. For the census, I use both the full census and just the Artibonite distribution to get the best comparison possible.
If we look at the highest bin, which is where the census median lies, the Artibonite census data has more plots there than the rest of the country, confirming that Artibonite plots are typically larger because of the higher agricultural productivity. But in the cadaster, this is the second-lowest bin. The modal plot is smaller than 0.32 ha.
What does it mean?
Since all of these data sets are non-representative samples of Haitian land, I’m slow to make a conclusion about Haitian land generally. But I’m surprised that the bias I expected still produced results smaller than expected.
But I will say that for the Artibonite cadaster and rental data, the sizes were confirmed by a surveyor; and for the irrigation data, people are paying for water. In all three datasets, there were incentives and mechanisms to report as accurately as possible. In the census data, however, people self-reported. If these data sets were representative samples, I would believe them more than the census.
So I cautiously say Haitian farms were smaller than previously thought. At most, the median was 1.29 ha (1 carreaux). But it’s possible that operating sizes were much smaller. Combine that with the lack of large-scale plantations, it’s no surprise that Haiti’s agriculture has lagged the region.