Border Transgressions

Guest Post by Willis Eschenbach

There is a new paper out in the Proceedings of the National Academy of Sciences called Linkages among climate change, crop yields and Mexico–US cross-border migration (hereinafter L2010). It has Supplementary Online Information (SOI) here. The editor of the paper is (the late) Dr. Stephen Schneider.

The paper basically advances the following theory of linkages:

Climate Change —> Reduced Mexican Crop Yields —> Migration to US

Hmmmm … their Abstract says:

Climate change is expected to cause mass human migration, including immigration across international borders. This study quantitatively examines the linkages among variations in climate, agricultural yields, and people’s migration responses by using an instrumental variables approach. Our method allows us to identify the relationship between crop yields and migration without explicitly controlling for all other confounding factors. Using state-level data from Mexico, we find a significant effect of climate-driven changes in crop yields on the rate of emigration to the United States. … Depending on the warming scenarios used and adaptation levels assumed, with other factors held constant, by approximately the year 2080, climate change is estimated to induce 1.4 to 6.7 million adult Mexicans (or 2% to 10% of the current population aged 15–65 y) to emigrate as a result of declines in agricultural productivity alone.

Although the results cannot be mechanically extrapolated to other areas and time periods, our findings are significant from a global perspective given that many regions, especially developing countries, are expected to experience significant declines in agricultural yields as a result of projected warming.

YIKES! … scary. Makes a man think seriously about mitigation.

I often divide things into the good, the bad, and the interesting. Regarding this study, first, the good. The authors have done a workmanlike job of pointing to the data that they used, all of which is online. This is to be highly commended, as it allows a quick determination of the validity of their work.

Next, the bad.

Because they were clear about their data, I was able to replicate their results exactly for the corn yields. My practice is to make replication the first step in any analysis of this type. It verifies whether they have done what they say they have done. In doing so, I discovered a most curious thing.

First, a small digression. “Yield” is how many tonnes of a crop are produced per hectare (or acre) harvested. Yield is affected by a number of things, including location, soil quality, and climate. If the yield in a certain location starts to fall, this is an indication that something is going wrong in the farming cycle in that location.

The curiosity that I discovered is that the paper calculates “yield” in a way that I had never seen. Yield is defined as how much crop production you get for every hectare (or acre) that was harvested. The authors, on the other hand, calculated yield as the amount produced for every hectare (or acre) that was planted. This often yields a very different number.

The source of their data is here.  Click on the “Maiz Grano” (Corn) in the first column, mid page. On the resulting page, click “Producción” (Production), second button from left. Then look in the far left column and click on the “Anuario” (Annual) button. Select 2004 as the year (“Año”) and press the “Consulta” button.

Now take a look at the data for 2004. The headings are:

Ubicación, Sup. Sembrada, Sup. Cosechada, Producción, Rendimiento

Or in English

Location, Area Planted (ha), Area Harvested (ha), Production (tonnes), Yield (tonnes/ha)

Over the period in question (1995-2004) Baja California averaged about 3 tonnes of corn per hectare. For Baja in 2004, their site says

BAJA CALIFORNIA, 592 hectares planted, 10 hectares harvested, 25 tonnes produced, yield 2.5 tonnes/ha

Note that, in common with other authorities, the Mexican web site itself calculated yield as production divided by area harvested, not divided by area planted. This is the normal definition of “yield” used by all other analyists. For example, from the UN Food and Agriculture Organization (FAO) web site glossary we have (emphasis mine):

Title: Crop yield

Definition:

Harvested production per unit of harvested area for crop products. In most of the cases yield data are not recorded but obtained by dividing the production data by the data on area harvested. …

“Harvested area” in turn is defined as:

Title: Area harvested

Definition:

Data refer to the area from which a crop is gathered. Area harvested, therefore, excludes the area from which, although sown or planted, there was no harvest due to damage, failure, etc. …

From this, it is clear that the authors of L2010 are not calculating the yield correctly. They have calculated the yield for Baja 2004 as 25 tonnes / 592 hectares planted = 0.04 tonnes/ha, a meaningless result. This is why yield is always calculated based on the area harvested, not based on the area planted. Obviously, something happened in Baja in 2004 that wiped out most of the corn crop. But for the remaining area, the yield was 25 tonnes / 10 hectares harvested = 2.5 tonnes/ha, not far from normal.

Overall, this is a very significant error. To take one example of the effect of the error, Figure 2 shows the correlations between Mexican annual temperatures and corn crop yields (correctly and incorrectly calculated).

Figure 2. State by state correlations between annual temperature and corn crop yields, 1995-2004. “Yield” is production / area harvested. “Incorrect Yield” is production / area planted, as used in L2010.

Note that in some States (Aguascalientes, Campeche, Yucatan), one dataset shows a very small correlation between temperature and yield, while the other shows 20%-40% correlation. In some cases (Nueva Leon, Queretaro, San Luis Potosi) one shows positive and one shows negative correlation. Overall, there are many results which are significantly different.

Because the correlations of the yield are central to their analysis, this error invalidates the paper and requires the recalculation of all the relationships. Remember that their thesis is:

Climate Change —> Reduced Mexican Crop Yields —> Migration to US

Note that there are two separate mathematical relationships in their claim. One relates climate change (temperature and rainfall) to changes in yield. The other relates changes in yield to migration rates. An error in the yield, therefore, requires a recalculation of both relationships, with new error bounds, etc.

Since the original web site is in Spanish, this error may simply be a misunderstanding of what the web site says. However, that slides over the question of why they didn’t simply use the yield figures provided in their data source …

I have posted up the Area Planted, Area Harvested, Production, Annual Temperature, and Yield figures here as an Excel spreadsheet. To determine which one they used (area planted or area harvested), it is necessary to take 5-year averages of the data (1995-1999 and 2000-2004) and compare the answers to Table S1 of the Supplementary Online Information. I can reproduce their results only by the incorrect usage of area planted instead of area harvested. Note that “Log Corn Yield” in Table S1 of their paper is the natural log (ln) of the yield.

I have pointed out some good about the study, and some bad, so onwards to the interesting. One interesting thing to me is the variety of responses of different states to increased or decreased temperatures. In a third of the Mexican states, warmer is better for corn (positive correlation). In two-thirds of the Mexican states, on the other hand, cooler is better for corn. Hmmm …

Another interesting thing is the change in the Mexican country average yield for corn. Figure 3 shows both the country average yield and average annual temperature for 1995-2005:

Figure 3. Mexican Corn Yield (red line, left scale) and Temperature (blue line, right scale) Photo Source

Fig. 3 highlights one of the real shortcomings of their study. This is the very short time period that they are investigating. However, taken at face value, this graph does not give much credence to the idea that increasing temperatures will reduce Mexican corn yield … (note that I make no claim that this relationship is meaningful or statistically significant. I only say it does not support the authors’ argument.)

As noted above, there are two mathematical relationships involved in their claim. One is temperature/precipitation vs yield, and the other is yield vs emigration. For the yield vs. emigration, the Mexican dataset is short. So I understand that they have to make do with what they have. But yield versus temperature has a much longer dataset. The temperatures from their source span 1971 to the present, and the state-by-state crop data goes back to 1980. So they should have established the corn yield/temperature link using all of the data available (1980-2009), even though the other yield/emigration link has so much less data.

How does something like this get published? I suspect that this is another example of a member of PNAS using their “Proceedings” publication as a vanity press with little in the way of peer review. The article is edited by Stephen Schneider, who also edited the other recent “blacklist” paper, so it’s clandestinely flying across the border under the peer-review radar …

Hopefully, this will be the last of the posthumous Schneider “science” for us to deal with. The only good thing about Schneider was that when I saw his name on something, I knew I could likely find errors in it … made my job that much easier.

Look, I don’t like to speak ill of the dead. Stephen Schneider was probably a nice man who loved his family and petted puppies and brought the homeless blankets and dinner. But his general claims were often a “post-normal science” abomination, and his scientific work (as in the present instance) was sometimes very slipshod.

In particular, Schneider is noted for his statement regarding the obligations of scientists:

To capture the public imagination, we [scientists] have to offer up some scary scenarios, make simplified dramatic statements and little mention of any doubts one might have. Each of us has to decide the right balance between being effective, and being honest. This ‘double ethical bind’ we frequently find ourselves in cannot be solved by any formula. Each of us has to decide what the right balance is between being effective and being honest. I hope that means being both.

To me, the most scary scenario is scientists who balance their honesty with effectiveness, or with anything else for that matter. I don’t want scientists who make little mention of their doubts. I don’t want scary scenarios from scientists, that’s why God made Hollywood and the BBC.

I want scientists who are as honest as possible, about their doubts and everything else. Schneider’s view, that scientists should balance honesty and effectiveness, is extremely and insidiously dangerous to science.

So, as un-PC as my view might be, I am overjoyed to see the last post-mortem gasp of Schneider’s apocalyptic alarmism. Am I glad he is dead? No way. As the poet said,

Each man’s death diminishes me,

For I am involved in mankind.

I am very happy, however, that he is no longer teaching at Stanford, that he is no longer writing garbage for me to wade through, and that he is no longer busily filling up the porches of the Stanford students’ ears with “cursed hebenon” …

My regards to all,

w.