Saturday, July 30, 2011

The relationship between hunger and petroleum consumption-Part 5

Here, I summarize parts 1-4 and put my results into the context of the broader ideas of peak oil, food production and population decline. 

A number of readers may think that it is obvious, or common knowledge, that food production and fossil fuel consumption are directly linked, and therefore, the higher the level of petroleum consumption in the world, in a region, or in a country, the greater the amount of food produced and the less hunger and starvation there will be.  The implications of reaching peak oil and declining along the down side are equally clear—less oil means less food, more hunger, and eventually starvation. 

This was definitely my thinking, until a few months ago when I started to look at the relationship between petroleum consumption and population growth for different regions of the world, and some individual countries, during periods when petroleum consumption dramatically declined.  To my surprise, the population often just kept increasing at the same rate or at most, the rate of increase slowed down a bit.  That is, I saw nothing that one would call a “die off” (see Estimating the End of Global Petroleum Exports; Part 7 and on).

I speculated that the reason why the population didn’t decline was that foreign food-aid has always stepped to substantially prevent what would otherwise be famine and death.   Because of this, I speculated that I needed to use a “softer” indicator of food production than population or population change.

In this series, I used three different “softer” indicators to assess a country's food production system.  That is, the ability of the system to prevent hunger, the percentage of the population with a high BMI and the Food Supply energy.  I looked at how these indicators related to the country’s petroleum consumption rate.  The results of my analysis suggest that regardless of which indicator I considered, a sharp decline in the indicator occurs when per capita consumption rates are less than about 1 barrel per person per year (b/py).   

I think that this could have important implications for which countries or regions of the world will be the first to have problems with food production and starvation, as petroleum production rates go into decline following peak oil.

Summary of Parts 1-4

Figure 8 shows a composite plot of the trend lines of the relationships between per capita petroleum consumption and the three indicators that I studied in Parts 1-4: the global hunger index (GHI, 2010 data); the percentage of the population with a BMI of greater than 25 (%BMI>25); and the available food supply energy (FS; 2007 data).   If you want to see the actually country by country data look at Figure 3, 5 and 6 in Parts 1-4.

The vertical scale represents different units depending on the indicator being considered.  For GHI and %BMI>25 I can use the same index scale and percent scale as reported in Parts1-2 and part 3, respectively.  For FS I converted the food supply energy into a percentage based on the assumption that the trend line values of food supply energy at 0.01 b/py and at 60 b/py corresponded to 0 percent and 100 percent, respectively. 

I think that Figure 8 nicely illustrates that the largest changes in the indicator occur at less than about 1 b/py.  That is, a per capita petroleum consumption rate of about 1 b/py represents an inflection point, below which the food production system, as represented by these indicators, falls apart.

I also looked at three or four subgroups within the data sets for each of these indicators to test whether or not there was an overall significant difference in the mean per capita petroleum consumption rates of the countries within each the subgroups.  A one-way analysis of variance indicated that the capita petroleum consumption rates of the three or four different food supply groups were not all equal (p<0.05). 

I continued the analysis of variance to make multiple group comparisons to see which of the subgroups per capita petroleum consumption rates were significantly different from each other.

Figure 9 summarizes the findings of these tests:


In general, the subgroups of countries with the lower food system indicator (i.e.; alarming or serious GHI; lowest percentage BMI>25; lowest Food Supply energy) have significantly lower mean per capita petroleum consumption rates than the subgroups of countries with the higher food system indicator.  Notice that all of the subgroups with lower food system indicator have mean per capita petroleum consumption rates ranging from about 0.6 to 0.8 b/py.  The subgroups with the second lowest indicator (for GHI and FS) have mean per capita petroleum consumption rates ranging from about 0.9 to 2.4 b/py. 

Finally, for the two indicators GHI and FS, I was able to compare the country-by-country changes in indicator from an earlier date (1990) to a more recent date (2010 for GHI and 2007 for FS) to see if the indicator and per capita petroleum consumption rates changed in a manner that is at least consistent with idea that petroleum consumption is needed for food production.

That is, if the global hunger index increased for a country from 1990 to 2010, did petroleum consumption go down, or, if the global hunger index decreased for a country from 1990 to 2010, did petroleum consumption go up?

Similarly, if the food supply energy for a country decreased from 1990 to 2007, did petroleum consumption go down, or, if the food supply energy for a country increased from 1990 to 2007, did petroleum consumption go up?

The answer to both questions is yes.  I did SIGNS tests for the 99 countries for which I had both GHI and petroleum consumption data in both years, and, for the 124 countries for which I had both FS and petroleum consumption data in both years.  The result of the test indicated that these proposed relationships occurred in more countries than expected by random chance (p<0.01).  That is, the change in global hunger index was inversely related to the change in per capita petroleum consumption rate and the change in food supply energy paralleled the change in change in per capita petroleum consumption rate. 

Conclusions—1 barrel per person per year

So, have I convinced you that about 1 barrel per person per year is a threshold amount of petroleum consumption needed to maintain the food production system?  

Well, I am at least 95% convinced (inside statistical joke) that there are significant difference in the per capita petroleum consumption rates in groups of countries with lower food system indicators versus groups of countries with higher indicator.  Those subgroups with lowest indicator all have mean per capita petroleum consumption rates in the range of 0.6 to 0.8 b/py.  The subgroups with the highest indicators range from 6 to 15 b/py.   This is consistent with the trend lines, which for all three indicators, there is a rapid decrease in the indicator as one transitions from countries with per capita petroleum consumption rates of greater than to less than about 1 b/py.

I am also 99% convinced (another inside statistical joke) that these food system indicators change in a manner that is consistent with petroleum consumption being a necessary ingredient to the food system.  This study doesn’t prove causation, which may be very hard to prove, but it does seem that increased petroleum consumption genrally means increased food production and decreased petroleum consumption generally means decreased food production.

Ho-hum you might say, I knew all of this before.  

Well, to the best of my knowledge no one has ever looked at this before, so I would be interested to know how you knew this.  May be you just assumed that food production and petroleum consumption are linked 1:1, as I once did. 

I think, however, that the relationship is more nuanced than that, and that comes back to my hypothesis that 1 barrel per person per year represents a critical inflection point, below which serious problems with the food production system occur. 

No, I do not see one barrel per person per year as some absolute limit that is written in stone—the data is just too scattered for that kind of conclusion.  

For instance, as illustrated in Figures 3 and 4 of part 2, many countries consuming as low as 1-3 b/py can still have a moderate to low global hunger index, although some other countries can have serious or alarming hunger index values.  Similarly, as illustrated in Figure 6 of part 4, the bulk of countries consuming 1-3 b/py have a Food Supply energy availability of more than 2230 kcal per person per day, although there are a few examples of countries where the Food Supply energy dips as low as 2000 kcal per person per day.

On the other hand, there are no countries that have a per capita consumption of less than 1 b/py and a global hunger index that is anything less than serious, for the two time periods studied.  Likewise, in the two periods studied, only one country with a per capita consumption of less than 1 b/py had a Food Supply energy that was greater than the overall country average Food Supply energy (China in 1990 and Ghana in 2007).

If I have convinced you that a consumption rates 1 b/py is a danger-zone, and you believe that we are at peak oil, then the implications should be clear to you: as petroleum consumption declines below about 1 barrel per person per year in individual countries, or regions, a crisis in food production will spread.  And, it will be the regions with consumption rates declining to the 1 b/py level that will likely have problems with their food system first.

As I already showed in my previous series (see Part 10, Figure 18) Africa is very close to 1 b/py right now, and, Asia-Pacific is not far behind.

   

Friday, July 22, 2011

The relationship between hunger and petroleum consumption-Part 4

Parts 1 and 2 looked at the relationship between the global hunger index (GHI) and per capita petroleum consumption, and, part 3 looked at the relationship between per capita petroleum consumption and BMI. 

Here in Part 4, I examine the relationship between the food supply energy for individual countries, as estimated Food and Agriculture Organization of the United Nations, and per capita petroleum consumption.

At the outset for new visitors, I feel compelled to say that this is not a dieting blog.  Also, I am not advocating for or against an all-bacon diet or an all-cracker diet, or any kind of diet.  I have no personal experience with either diet—although apparently there are people in America at least who have tried these diets (sad, sadder, and very sad).  If you want to talk about these or other types of diets, this is not the place to be.

Rather, I am interested in the role that petroleum plays in food production.  I am interested in this because I think that we are in the midst of peak oil, where petroleum production declines, and therefore petroleum consumption must also decline.   I think that this will have negative implications for the world’s petroleum-driven food production system.  That is, there will be problems in feeding the world’s growing population as  petroluem consumption declines over the next 10-30 years.  My assumption is that large numbers of people will die of starvation if/when the food production system is deprived of petroleum below a certain critical level.

What I am particularly interested in is trying pin-point at what point the petroleum consumption rate become too low to support the production of food in sufficient quantities to support a country’s population, and, people start to get hungry and then start to die of starvation. 

Data sources and my selection criterion

If you are a food junky, statistical speaking that is, then the Food and Agriculture Organization (FAO) is the place to go pig out.  The FAO’s compilation of statistical data is distributed over several different divisions, and, the data base is simply enormous.  I have spent most of my free time recently at FAOSTAT, the FAO Statistical Database, which contains, “over 1 million time-series records from over 210 countries and territories covering statistics on agriculture, nutrition, fisheries, forestry, food aid, land use and population.” 

I’m telling you reader, give up watching the “Bachelor,” or whatever, FAOSTAT has endless hours of entertainment, if you are only willing to work for it! 

But, I digress...

Among the legions of statistics complied at FAOSTAT are statistics for the Food Supply, defined as:
total and per caput food supplies available for human consumption during the reference period in terms of quantity and, by applying appropriate food composition factors for all primary and processed products, also in terms of caloric value and protein and fat content. Calorie supplies are reported in kilocalories. The traditional unit of calories is being retained for the time being until the proposed kilojoule gains wider acceptance and understanding (1 calorie = 4.19 kilojoules). Per caput supplies in terms of product weight are derived from the total supplies available for human consumption (i.e. Food) by dividing the quantities of Food by the total population actually partaking of the food supplies during the reference period, i.e. the present in-area (de facto) population within the present geographical boundaries of the country.
from, Food Supply Notes (emphasis added)

I added the emphasis to make clear that the FAO’s definition of “Food Supply” is the amount of food that is available for human consumption. This amount is not necessarily equal to the food that is consumed by humans.  I expect that different countries (e.g., USA compared to Ethiopia) waste their food supply to different degrees (e.g., the USA more than Ethiopia), but my main interest is in the total food available for consumption, as this is likely to reflect the petroleum consumed in making the food. 

As a side note, the FAO’s “Food Supply” omits the food supply to feed livestock for subsequent human consumption.  It is possible to include this because “Feed” is listed as a separate category in FAO’s data bases, although it is not reported in energy units.   However, to correct for feed for each of the +100 countries considered here would take a substantial amount of work that I am not willing to take on at this time. 

Food Supply is given in kilocalories/capita/day (kcal/pd), which is calculated, as noted above, by applying appropriate food composition factors for all primary and processed products, in terms of caloric value.  Expressing food in terms of caloric value is a way of allowing all of the different food types consumed to be compared on a same scale.  Yes, yes, it would be nice to have some measured of the quality of the food supply, but I am unaware of such a unitary measure being available for all countries and all years.  So, we are stuck with caloric value for the time being.

FAOSTAT makes estimates of “Food Supply” for every year and nearly every country in the world since 1961—did I mention, hours of entertainment?  

To make my analysis more tractable, and comparable to the analysis I did in Parts 1 and 2, I decided to focus only on 1990 and the most recent data reporting period of 2007. 

Once again, I used per capita petroleum consumption rate (barrels per person per year, b/py), calculated as described in part 1 (from the EIA and US census bureau), for the two years, 1990 and 2007, of interest. 

Also, in addition to considering only those countries for which all both food supply and per capita petroleum consumption data are available (about 175 countries in 2007) I decided to omit all those countries (about 30 countries in 2007) that had a population of less than 1 million people in 2007.  For the most part, this resulted in the omission of a few dozen small island-states and a few city-states, whose food supply system I beleive might not be representative of the bulk of the world’s population.  I could have tried some type of weighting scheme, but simply providing a cut-off was the most expedient way to biasing away from these small states.  That is, that a state with a population of a few 10s of thousands of people should not have the same weighting as state with 10s, 100s, 1000s of millions of population.  This is a somewhat arbitrary cutoff on my part, but, the remaining countries data considered here still accounted for over 98% of the world’s population for the two years considered.

Results and Statistical Analysis

Figure 6 presents a plot of Food Supply (in units of kcal/pd) versus per capita petroleum (b/py) for all 145 countries for which a food supply estimate was made in 2007 and otherwise meeting my criterion: 

As in my previous posts in this series, the solid curve in Figure 6 corresponds to a power equation trend line also shown in the figure.  And, again, as in parts 1-3, I don’t ascribe any particular meaning to the power curve or its best-fit parameter values—but the curve does help show the general trend in the relationship between food supply and per capita petroleum consumption.

Figure 7 shows the analogous plot for 1990:

As illustrated in Figures 6 and 7, the trend is for there to be lower per capita consumption levels for those countries where the available per capita food supply energy is lower than in other countries. 

The average food supply and standard deviation for 1990 and 2007 equaled 2571±573 (n=125) and 2759±521 (n=145), respectively. 

Based on these averages and their standard deviations, I decided to define four different food supply groups, corresponding to the group averages plus or minus one or two standard deviation (sd) values.  The four food supply groups and their corresponding average(±sd) per capita petroleum consumptions, and, the number of countries in each group (n), are presented in the Table below:


Mean per capita petroleum consumption for groups having different degrees available food supply
Food Supply Group
Range of Food Energy Available (kcal/pd)
Per capita petroleum consumption (b/py)
Year =>
1990
2007
1990
2007
G1: Low
< 1998
< 2238
0.56±0.57 (n=21)
0.67±0.71 (n=26)
G2: Moderate
1998-2571
2238-2759
1.8±2.8 (n=49)
2.4±2.7 (n=46)
G3: Good
2572-3144
2760-3280
7.3±6.3 (n=26)
9.4±9.6 (n=50)
G4: High
> 3144
> 3280
14.8±8.7 (n=28)
11.6±6.5 (n=23)


There are a lesser number of countries in 1990 (n=124) compared to 2007 (n=145) mainly due to the birth, or re-birth, of several new countries after the dissolution of the Soviet Union after 1990, and, because some countries had no reported petroleum consumption in 1990, but did report petroleum consumption in 2007.

If the amounts of food energy available in the above table seem high, remember that this is food available for human consumption, and, not the actual amount consumed.   For instance, the FAO also periodical reports a “Minimum Dietary Energy Requirement” for about these same countries. 

The Minimum Dietary Energy Requirement estimated by the FAO tries to account for country-by-country differences in gender and age structure, and other factors, which in turn affects energy requirements for the population.  

For all countries reported, the average Minimum Dietary Energy Requirement equaled 1790±82 kcal/pd for 1990-92 and 1825±82 kcal/pd for 2004-06 (the latest reporting period). These are pretty tight standard deviations, in my opinion, which in turn suggests that for the most part, all populations need about the same minimum amount of food, regardless of which country we are considering.

It makes sense that the food supply energy available would be much higher than minimum food energy required.  Since the food supply within each country is not distributed evenly, if the food supply available was ever close to the minimum food needed, then that would mean that a major fraction of the population would be starving. 

The FAO uses these two numbers, and an ssumed or derived distribution curve (f(x)), to estimate the percentage of a population that is undernourished, such as generically indicated below:


The relative area under the curve on the left-hand-side (red arrow) of the distribution curve corresponds to FAO’s estimate percentage of population that is undernourished in a particular hypothetical population. 

The FAO’s percentage undernourished is actually one of the three hunger-related indicators (i.e., PUN, the proportion of the population that is undernourished) that makes up the GHI, reported by the International Food Policy Research Institute, as discussed in Parts 1 and 2. 

ANOVA of the four groups with different food supply
The four different Food Supply groups tested, G1-G4 were defined in the above table.

I started off with a one-way analysis of variance (ANOVA) to test the hypothesis: are the mean per capita petroleum consumption rates of the four groups all equal (i.e., G1=G2=G3=G4).  This hypothesis was rejected at p<0.05.  That is, the per capita petroleum consumption rates of the four different food supply groups are not all equal. 

A subsequent multiple group comparison for the 2007 data, using the Tukey test (p=0.05), revealed that the per capita petroleum consumption rates between groups significantly differed in following respects:
1) G1 is different than G4 and G3;
2) G2 is different than G4 and G3;
3) G3 is not different than G4; and
4) G2 is not different than G1.

A subsequent multiple group comparison for the 1990 data, using the Tukey test (p=0.05), revealed that the per capita petroleum consumption rates between groups significantly differed in following respects:
1) G1 is different than G4 and G3;
2) G2 is different than G4 and G3;
3) G3 is different than G4; and
4) G2 is not different than G1.

Statisticians like to summarize such finding graphically, where the continuous line represents group means that are not statistically different, and, a break in the line means a significant group difference exists:

Is the change in food supply from 1990 to 2007 consistent with the expected change in per capita petroleum consumption?

Once again I have matched pairs of Food Supply data for a number of countries (n=124).  This allows me to test to see if changes in the Food Supply from 1990 to 2007 are consistent with my expectations of how the per capita petroleum consumption should change.  That is, if the Food Supply when up from 1990 to 2007, did the per capita petroleum consumption also go up?  Conversely, if the Food Supply when down from 1990 to 2007, did the per capita petroleum consumption also go down?  This is what I would expect if Food Supply and per capita petroleum consumption are causally linked.  Or, at least if the direction of change in these two parameters are random, then it is harder to see how there could be a causal link.

To test this hypothesis, I once again did a SIGNS test similar to that described in Part 2 of this series.

To set up the data properly, for the 124 countries for which we have data in both 1990 and 2007, if Food Supply increased and per capita consumption increased, or, if the Food Supply decreased and per capita consumption decreased, then I assigned a value of "1."  For the opposite scenario, if the Food Supply increased and per capita consumption decreased, or, if the Food Supply decreased and per capita consumption increased, then I assigned a value of "0."

The SIGN test is a test of the null hypothesis.  That is, the hypothesis is that probability of assigning a value of "1" and "0" for a particular country are equal.  That is, if the proposed parallel relationship between Food Supply and per capita consumption is not true, then there should be an equal numbers of 1s and 0s (i.e., 62, in either group for a population of 124 countries).

My analysis showed that, of the 124 countries with paired data, 79 followed the proposed relationship and 45 did not.  The lower and upper criterion values (i.e., a two-tailed SIGNS test) at a p < 0.01 are 48 and 76, respectively, and since 79 is outside of this range, the null hypothesis is rejected. 

In other words, there is a less than 1% chance that proposed direction of change in Food Supply with an parallel direction of change in per capita consumption is due to random chance alone. 

I took this analysis one step further, and repeated the SIGNS test of the null hypothesis for two subgroups.

The first subgroup was all the countries (n=59) with a Food Supply of greater than the average 2759 (i.e., the mean Food Supply in kcal/p/d for the 124 countries considered here) in 2007; the second subgroup was all countries (n=65) with a Food Supply of 2759 or lower (i.e., the average or below average food supply).

The null hypothesis was rejected for subgroup of countries with the above average Food Supply at p<0.01.  For subgroup of countries with the average or below Food Supply, the null hypothesis was rejected—but only at p<0.1.  

This suggests that the expected parallel relationship between the Food Supply energy per capita and per capita petroleum consumption holds for both higher and lower levels of food supply, although the level of uncertainty is greater for the later than the former.

Conclusions

As illustrated in Figures 6 and 7, regardless of the year studied, there is a rapid decrease in per capita petroleum consumption as we look at countries with high to good food supply energy per capita, to countries with a moderate to low food supply energy per capita.

The trend lines basically show the same trends for both 1990 and 2007—the lower the food supply per capita, the lower the per capita petroleum consumption, with a sharp transition occurring at about 1 barrel per year per person.

There are statistically significant differences in the per capita petroleum consumption rates between the two lowest food supply groups (G1 and G2) and the two highest food supply groups (G2 and G4).  And, for 1990, the second highest (G3) and highest (G4) food supply groups had significantly different per capita petroleum consumption rates.

These results are similar to what I observed in Parts 1-3, using two different types of hunger measures (GHI and BMI). 

What does this all mean? 

I will offer my summary and thoughts in the next, and final installment of this series.

Saturday, July 9, 2011

The relationship between hunger and petroleum consumption-Part 3

Parts 1 and 2 looked at the relationship between the global hunger index (GHI) and per capita petroleum consumption.  Here in part 3, I describe the relationship between per capita petroleum consumption and another potential indicator of hunger—body mass index (BMI). 

Most people are familiar with BMI, not as a measure of hunger, but rather quite the opposite—an over-abundance of food leading to unhealthy levels of weight gain.   Roughly speaking, or speaking roughly, adults with a BMI of 25 of more, are considered to be fat or “overweight” while a BMI of 18.5 or less is considered thin or “underweight.”

My idea, or hope, was that there would be a broad range of BMI measurements reported for both developed and undeveloped countries, and, this would allow me to use BMI as an alternative indicator of “hunger.”  That is, if petroleum consumption is important to food production, then countries with high per capita petroleum consumption rate would have a higher percentage of the population in the “overweight” category (BMI > 25) than countries with a low per capita petroleum consumption rate.  Or, countries with a low per capita petroleum consumption rate would have a higher percentage of people in the “underweight” category (BMI< 18.5) than countries with a hign per capita petroleum consumption rate.

Data sources and my selection criterion
The best source of accumulated BMI data that I could find was at the World Health Organization (WHO), which makes available, in spreadsheet form, BMI measurements for a large number countries at various times collected over the last 30 years.  I use this data, plus per capita petroleum consumption rate (barrels per person per year, b/py) calculated as described in part 1 (using EIA and US census bureau data) for the particular year in which the BMI data was reported. 

To make the comparison between countries as uniform as possible, I limited my analysis to the latest data for only those countries that reported combined national data that included both sexes and both urban and rural areas.

To my disappointment, there was a relative lack of data reported for those countries identified in parts 1 and 2 as having a low GHI.  For instance, of the 58 countries identified as having serious hunger (i.e. GHI > 10), only 13 had reported BMI data that met my criterion.  So I lost 40 countries in an important range of interest.  I also lost another 29 countries which had moderate or low report hunger values (GHI<10) but no BMI data.

I did, however, pick up data for 32 countries in EU, NA, AP and ME for which there was BMI data but no GHI data.  Unfortunately, all of these pick-ups were developed countries with a high per capita petroleum consumption rate. 

Additionally, I found that for many countries, the WHO data base did not have BMI statistics for “underweight” and meeting my requirements (national, both sexes, rural and urban). Rather, the most common statistic was only data for “overweight” (BMI>25) and somtimes “obese” (BMI>30).   Therefore, I restricted my analysis to only considering the percentage of the population having a BMI>25. 

Another problem is that the BMI is not reported or collected in the same year for every country.  In fact, the reports of BMI that would meet my criterion, ranged anywhere from 1988 to 2009, although most of the data is from the 2000s.  Besides adding an undesired variable into the data (time), this made it difficult to calculate the per capita petroleum consumption because it meant I had to get the population and national petroluem consumption figures for a variety of different years for each country. 

Results
Figure 5 presents a plot of the percent population having a BMI>25, versus per capita petroleum consumption rate, for the 81 countries that met my selection criterion.   

Once again I used a power equation, the solid line in Figure 5, to help show the general trend in the data.  I don’t ascribe any particular meaning to the power curve or its best-fit parameter values—but it does help illustrate the general trend.  According to the trend line, at a per capita petorleum consumption of 1 b/py, the percentage of overweight people is down to 23% and that percentage doubles to 46% by the time consumption is at 9 b/py.

Of the 13 hold-over countries from Parts 1 and 2 which had a GHI of greater than 10 (serious or alarming/extreme hunger) and a reported BMI, the average percent of the population overweight was only 18%±12%.

As shown in Figure 5, there were only 13 countries for which less than 30% of the population was overweight (i.e., BMI >25).  Eleven of those countries also had a per capita consumption rate of less than 2 b/py.  Two strong outliers were Japan (JP) and Singapore (SG).  It may be that a BMI of greater than 25 is not a realistic definition of over-weight for Japan and Singapore, as these two countries have reduced their cut-off for being overweight to BMI>23 (ref 13 and 14 in Body Mass Index). 

For the 19 European countries that were added to the data base (i.e., countries that don’t have a reported GHI, but do have a reported BMI) the average per capita petroleum consumption was 13 b/py and the percentage of the population that was overweight was on average 50% (variable years). 

New Zealand had a slightly higher per capita petroleum consumption at 14 b/py and a higher percentage of overweight people at 63% (2006).  Australia’s petroleum consumption was higher still, at 16 b/py, but “only” 49% of the population was overweight (2004).

South Korea (KR) had a still higher per capita consumption of 17 b/py, but only 33 % of the population has  a BMI>25 (2007), but maybe the cut-off for overweight should be adjusted for this country, similar to Japan and Singapore (see AMA Commentary by Gallagher).

The USA (US) and Canada (CA), with per capita consumption rate of 25 and 26 b/py (2007 and 2004, respectively), had an even higher percentages of overweight people at 68 and 59%, respectively. 

One African country actually makes into the ranks of the high per capita petroleum consumers, 25 b/py, and a correspondingly high population that is overweight 61% (2004)— Seychelles (SY).

Lastly, there are 3 countries in the Middle East with high per capita petroleum consumption rates and very high percentages of the population being overweight:
Saudi Arabia (SA), consumption at 25 b/py and 72% overweight in 1997,
United Arab Emirates (UA), consumption at 37 b/py and 64% overweight in 2000, and
Kuwait (KW), consumption 51 b/py and 75% overweight in 2006.

ANOVA of three groups with different BMI percentages
I decided to omit South Korea, Japan and Singapore, because it seems that the global criterion for being overweight (BMI>25) is not applicable to these countries, and I don’t know how to make a correction for this. I divided the remaining countries into three groups:

G1: Low percentage overweight—30% or less overweight
G2: Intermediate percentage overweight—greater than 30% but less than 50% overweight
G3: High percentage overweight—greater than 50% overweight

A one-way analysis of variance (ANOVA) to test the hypothesis: are the mean per capita petroleum consumption rates of the three groups all equal (i.e., G1=G2=G3), was rejected at p<0.05.  That is, the per capita petroleum consumption rates of the three BMI groups are not all equal. 

A subsequent multiple group comparison, using the Tukey test (p=0.05), revealed that the per capita petroleum consumption rates of G1 versus G3 or G1 versus G3 were significantly different form each other, but G3 versus G2 were not significantly different.

Here is a summary of the average and standard deviation of the per capita petroleum consumption rates for the three groups:
Average and standard deviations of per capita petroleum consumption (b/py)
G1: Low percentage overweight (30<%)
0.80±.53*
G2: Intermediate percentage overweight (30>%<50)
8.9±6.6
G3: High percentage overweight (%>50
12±11

* significantly different than G2 and G3

Conclusions
I am a bit disappointed that there was not more data for countries with food and hunger problems, but, I suppose I should not be too surprised by this.  BMI was designed to quantify “rich-country” diseases associated with being too fat.  Countries with food and hunger problems are not that interested in measuring BMI.  Even moderately well-off countries don’t collect BMI statistics, it seems. 

Despite the limitations in the database, the trend is consistent with what I would expect for a petroleum-driven food production system not consuming enough petroleum to produce adequate food—a lower percentage of overweight people. 

The percentage of overweight people in a population is an inverse indicator of hunger.  Consequently, the trend line in Figure 5 is the inverse of the trend lines in Figure 1 and 2 in part 1.  The point where the percentage of fat people in a country begins to sharply decline occurs around a per capita consumption rate of about 1-2 b/py.  This is consistent with the analysis in parts 1 and 2 using the global hunger index (GHI) as my indicator of compromised food production where the GHI sharply increases when the petroleum consumption rate drops to about 1 b/py.

If there is a link between high BMI levels and increased risks of negative health consequences like atherosclerosis, heart disease, diabetes etc..., then maybe there can be a certain high rate of petroleum consumption that is actually harmful to one’s health. A population where 50 percent or more of the population is overweight doesn’t sound too healthy to me, and, this appears to be related to a consumption rate of about 11 barrels per person per year or higher.