Tuesday, June 28, 2011

The relationship between hunger and petroleum consumption-Part 2

In part 1, I presented my hypothesis that countries with a high hunger index level will have a low per capita petroleum consumption rate, and, countries with a low hunger index level will have a high per capita petroleum consumption rate.  The underlying idea is that below a certain critical level of oil consumption, the modern petroleum-driven food production system is unable to keep the population fed and this will manifest as a high hunger index.  The expected consequent large-scale famine might be mitigated by foreign food aid, but that food aid may not be enough to mitigate hunger.  That is, an elevated hunger index is a soft indicator of potential famine.

Here, in part 2, I delve deeper into the analysis of the data base of countries for which the International Food Policy Research Institute generated a Global Hunger Index, and, my estimates of those country's capita petroleum consumption rate.  My analysis suggests that there is indeed a significant link between per capita petroleum consumption and the Global Hunger Index, and, that a drop in petroleum consumption below one barrel of oil per person per year (b/py) correlates with serious or higher hunger categories.

Expanding the Global Hunger Index (GHI) data base

One issue I have with the GHI, as publicized by the International Food Policy Research Institute (IFPRI), is that any country with a low hunger index is simply reported as having a GHI of less than 5 (GHI< 5).  I represented those countries, in my plots shown in Figure 1 and 2 in Part 1, as having an GHI value of 2.5. 

I understand that countries with a low GHI are not the IFPRI’s primary interest and so they don’t focus their attention on these countries.   But for me, there is potentially important information locked up in that GHI < 5, especially when trying to compare changing trends in hunger index with changing trends in petroleum consumption. 

It dawned on me that I could calculate these GHI values myself because the IFPRI presented, in spreadsheet form, the underlying data that the GHI for every country is based on, including those countries with GHI < 5. 

Specifically, GHI is calculated according to the follow formula:
GHI = (PUN + CUW + CM)/3
where:
PUN=proportion of the population that is undernourished (in %)
CUW=prevalence of underweight children under five (in %)
CM=proportion of children dying before the age of five (in %)
The PUN, CUW and CM values for each country are reported in a spread sheet, which I used to calculate the explicit GHI values for surveyed countries reported by the IFPRI to have a GHI < 5.   This added explicit GHI values for 9 countries in the 1990 data base, and for 38 countries in the 2010 data base.

Using the revised data sets, the trend lines, used in part 1 to predict thresholds for transitioning from one hunger category to another, were only slightly altered—the only moderate difference is for the transition from low to moderate hunger index for 1990 (13.9 instead of 12 b/py):

Per capita petroleum consumption at transition points between hunger categories (revised)
http://crash-watcher.blogspot.com/
Hunger category transition
per cap consumption (1990)
per cap consumption (2010)
low => moderate
13.9
4.2
moderate  => serious
2.6
1.2
serious => alarming
0.50
0.33
alarming => extremely alarming
0.19
0.16

Regional analysis of GHI versus per capita petroleum consumption

I reordered the data for the each of the surveyed countries into seven regional categories that will be familiar to previous readers of this blog: the Middle East (ME), Africa (AF), North America (NA), South America (SA), Asia-Pacific (AP), Europe (EU) and the former Soviet Union (FS). 

These are the regional categories that I used or derived from the data presented in the BP Statistical Review, and, analyzed in my previous series, Estimating the End of Global Petroleum Exports.

Figures 3 and 4 are repeats of Figures 1 and 2 from part 1, except that I have used different colors to represent the data for each country within these seven different regions.  Additionally, I have expanded the horizontal scale, for per capita consumption, from 0 to 10 b/py, so that I could better see the data points in the most important region of these plots.  This causes the data for only four countries to not be displayed: Saudi Arabia, Kuwait, Libya and Panama.  However, these countries were still included in the trend lines and statistical analysis described below.


Also presented in Figures 3 and 4, are my explicitly calculated GHI value for those countries where GHI < 5, and the revised trend lines fit to the data sets with these explicitly calculated GHI values included. 

My impression of the regional data is that many African countries (orange circles) tend to cluster in the high GHI-low per capita consumption area of the plots (i.e., upper left hand corner).  Countries from Europe (dark blue), the former Soviet Union (pink), and Mexico (dark green, the only country from North America included in the survey), are clustered in the low hunger high consumption area of the plot (i.e., lower right hand corner) .  Countries from the Middle East (brown), Asia-Pacific (turquoise blue) and South America (lime green) are uniformly distributed in both of these areas.  There is no 1990 data for the 15 former Soviet Union countries and 6 European countries.

Another noteworthy feature is that there was no country for either of the 2010 or the 1990 data sets for which per capita petroleum consumption was less than 1 b/py and the hunger index was not greater than 10 (i.e., “serious” hunger or higher)—I depicted this area as a yellow box in the lower left corner of Figures 3 and 4.  I think that this result suggests that the petroleum consumption level of 1 b/py represents a critical threshold that correlates with a country’s loss in ability to feed itself at least well enough to prevent a "serious" hunger index value.

A Brief Survey of outliers

Although the data from 99 countries for 1990 and from 122 countries for 2010 for seven different world regions all generally follow about the same trend lines, there are a few prominent outliers that I want to spend a few minutes considering.  These are three countires with relatively high petroleum consumption countries but also s high hunger index: Angola (AO), Yemen (YE), Djiboti (DJ).

Angola

Compared to other countries with similar consumption levels (1.1 and 2.0 b/py for 1990 and 2009 respectively), Angola has a very high GHI that puts it in the “extremely alarming to alarming” categories designated by the IFPRI (40.6 and 27.2 for 1990 and 2010, respectively). 

In 1990 Angola was in the mist of its 27-year long civil war which ended in 2002, so this helps explains the higher-than-expected GHI.  Significant numbers refugees returning from neighboring countries probable still contribute to the high GHI in 2010.  Certainly, the trend of GHI decreasing from 40.6 to 27.2 and petroleum consumption to increasing from 1.1 to 2 b/py is consistent with the I would expect if petroleum consumption was important to food production.  Nevertheless, as Angola rebuilds from civil war there is still significant government corruption and cronyism which I suspect results in major portions of the population remaining in poverty and hunger. 

Apparently, there are wide disparities in the wealth of people in Angola, despite the fact that Angola is a major oil exported and a so-called “darling” of the oil company investment (Angola, one of the poorest places on Earth, is an oil industry darling).  My previous assessment of Angola’s petroleum production and consumption trends suggests that Angola’s consumption rate is increasing at about 8.4 percent per year and is on trend to increase its petroleum production by 30% over the next few years, after which it will likely hit peak oil and start to decline again.

I thought that this 2006 announcement from the World Food Programme (WFP) was pertinent:

WFP has said that a lack of funding had prompted it to wind down all its food aid operations in Angola by the end of the year, after three decades of direct involvement in the country.
....
When the war ended in 2002, WFP assisted with the long process of reconstruction and repatriation of Angolan refugees, which is still continuing.
More than 80,000 refugees are expected to return home from camps in Zambia, Namibia and the Democratic Republic of Congo.
In addition to the WFP food stocks currently in Lobito and Luanda, another 3,800 tons of food are due to arrive in the country shortly.

The cessation of food aid from the WFP in 2006 might also help explain why Angola’s hunger index didn’t go down more by 2010 compared to 1990.  Once Angola hits peak oil in the mid-2010s I wonder if food aid will start up again, or, if a lack of funding will result in famine.

Yemen

Yemen has even higher petroleum consumption levels than Angola (2.2 and 2.6 b/py for 1990 and 2009, respectively), but comparably very high hunger index (GHI 30.1 and 27.2, respectively).  Again, the trend from 1990 to 2010 is in the expected direction with the hunger index going down as petroleum consumption levels go up.  But still, as illustrated in Figures 3 and 4, most other countries with this level of consumption are at the borderline of moderate-to-serious hunger indexes.  What is going on here?

It may well be that Yemen is facing a different type of resource constraint on food production—water:

Despite its oil reserves, which provide about 90 per cent of the country's exports, high levels of poverty (40 per cent) and unemployment (35 per cent) make Yemen the only low-income country in the Middle East. At the same time, Yemen's oil reserves, on which its economy currently depends, are predicted to run out within the next ten years.

Water is another finite resource that requires urgent attention: with groundwater resources rapidly depleting, Yemen's greatest environmental challenge is water scarcity. Yemen already has one of the lowest rates of per capita water availability in the world, estimated at about two per cent of the world average. Due to the rapid depletion of groundwater resources, the water table is falling by about two metres per year, constraining agricultural production and causing chronic water shortages. Periodic droughts and desertification, coupled with a shift from food production to cultivation of cash crops, are also impacting agriculture.
...
Water scarcity is Yemen's main constraint to food production. Almost 90 per cent of water use is for agriculture, and a large proportion of this is due to inefficient irrigation techniques and the expansion of qat cultivation, which alone accounts for 30 per cent of water use. Six times more profitable than most food crops, and relatively easy to cultivate, qat cultivation has expanded at the expense of food crops, contributing to the dependence on food imports. At the same time, the mildly narcotic leaves account for up to 30 per cent of household expenditure, ranking second only to food for many poor people.
....
Water is being extracted from the Sana'a basin four times quicker than it is being replenished and, with a population growth rate of seven per cent, Sana'a could become the first capital city to run out of water. In response to growing water scarcity, the UK's Department for International Development (DFID) has been funding water harvesting projects in Sa'adah province where, in the village of Al-Qatab, a 4,300 cubic metre capacity cistern, hand-pump and precipitation tank have provided this mountain-top community with a reliable source of water.

(“Qat,”or Khat, is classified by the World Health Organization as a drug of abuse that can produce mild to moderate psychological dependence.)

The above excerpts from the New Agriculturist mentions the declining petroleum reserves in Yemen.   Even more serious, in my opinion, is that Yemen’s net petroleum exports is rapidly headed towards zero, as I illustrated in a previous article, Survey of Oil Exports from the Middle East.  As net exports hit zero, probably in the next few years, Yemen will lose its petroleum export sales as a source of income with which to buy food, making the prospects of food shortages and hunger even worse than present.

Djiboti

With per capita petroleum consumption at 7.7 and 6.0 b/py in 1990 and 2009, respectively, but a hunger index in the “alarming category,” Djiboti is the biggest outlier of all.  Like Yemen, food production in Djiboti seems to be constrained by an unfavorable climate with low rainfalls.  Djiboti is a Delaware-sized country with population of 820,000 and with 70% of the population living in the capital city (see World Food Programme, Djiboti).   Djiboti has little natural resources or industry, and its main economy appears to be serving as an import/export port for landlocked Ethiopia.  Unemployment is an incredible high 60%, and the population is heavily dependent on food imports (see Economy Watch, Djibouti Economic Statistics and Indicators).  My hunch is that most of the country’s petroleum consumption is related to its port activity and has little to nothing to do with petroleum use for food production. 


Is per capita petroleum consumption significantly different between different hunger index groups?

To answer this question, I divided the surveyed countries into four different hunger index groups of approximately equal sizes: G1, low hunger (GHI < 5); G2, moderate hunger (5 < GHI < 10); G3, serious hunger (10 < GHI 20); and G4, alarming/extreme hunger (GHI > 20).

The starting point of my analysis was a one-way analysis of variance (ANOVA) to test the null hypothesis: are the mean per capita petroleum consumption rates of the four groups all equal (i.e., G1=G2=G3=G4)? 

EXCEL 2003 has the capability of running a one-way (or single factor) ANOVA automatically (interested readers can see examples of this here or here).

Here’s the output summary from my analysis of the 2010 data set.
SUMMARY






Groups
Count
Sum
Average
Variance


G 1
38
286.26074
7.533177
91.626616


G 2
25
119.81623
4.792649
9.4168502


G 3
30
33.545026
1.118168
0.8135078


G 4
28
15.487953
0.553141
0.323333


ANOVA






Source of Variation
SS
df
MS
F
P-value
F crit
Between Groups
1064.989
3
354.9962
11.383976
1.32E-06
2.682132
Within Groups
3648.511
117
31.18385










Total
4713.5
120





The Average here is referring to the average per capita petroleum consumption for each of the four different hunger index groups.

Since the F statistic (11.2) is greater than the critical F value (2.68) I can reject the null hypothesis.  That is, the per capita petroleum consumption rates of the four hunger groups for 2010 are not all equal. 

I got similar results for the 1990 data set— the per capita petroleum consumption rates of the four hunger groups for 1990 are not all equal. 

The one-way ANOVA suggests that there are statistically significant differences somewhere in the data as a whole, but it does not pin-point exactly where those differences lie.  Putting it another way, if we are willing to think in terms of a causative relationship— the per capita petroleum consumption rate does appear to affect the hunger index.  But this analysis does not indicate whether all, or only some, levels of per capita petroleum consumption rate affect all of the different respective hunger index groups.

The answer that question, I performed a Tukey “Multiple Comparison” test on these same two groups of data from 1990 and 2010. 

Unfortunately, EXCEL does not provide a tool for doing this automatically, so I had to improvise and setup a means of doing the test within EXCEL. 

This is not hard to do, if one has a good example to work with.  In that regard, I used example 12.2 from Zar’s book, Biostatistical Analysis (2nd ed).  I love Zar’s book precisely because it provides many practical worked examples of many different types of statistical tests.  I highly recommend it. Biostatistical Analysis is now in its 5th edition.

Using the results from EXCEL’s one-way ANOVA as a starting point, I sorted the resulting averages for G1 to G4 in ascending order, and then applied the Tukey multi-comparison test to look for significant differences between the group means.  In each case, the null hypothesis being tested is that the average per capita petroleum consumption rates between the two hunger groups under consideration are not different.  Therefore, rejecting the hypothesis implies that the average per capita consumption rates are significantly different for the two hunger index groups being considered. 

(July 9, 2011: I detected an error in my calculation of the SE and therefore in q; consequently the statistical conclusions are somewhat altered in that the null hypothesis is no rejected for as many of the group comparisions; text to this section is revised accordingly. I feel better now, because for about a week the statistical conclusion did not jive with my intuition above the averages and SD, but I could not find the discrepancy).

Here are the explicit results for the 2010 data set:
Group
comparison
difference between means
SE (sqrt(ms/2((1/n1+1/n2))
q
crit q
(0.05,df,k)
Conclusion
G1 to G4
6.790678
0.973362
6.976518
3.685
reject
G1 to G3
6.41501
0.9641172
6.653766
3.685
reject
G1 to G2
2.7405282
1.0165676
2.695864
3.685
accept
G2 to G4
4.0501498
1.0773472
3.759373
3.685
reject
G2 to G3
3.6744817
1.069002
3.437301
3.685
accept
G3 to G4
0.3756681
1.0280025
0.365435
3.685
accept

The null hypothesis is rejected, for the comparisons between G1 and G4, G1 and G3, and G2 and G4, and accepted for the rest, at least at the 95 percent confidence interval (p<0.05).

That means that the both the serious (G3) and alarming/extreme (G4) hunger index groups have significantly lower average per capita petroleum consumption rates than the low hunger  index group.  Also, the alarming/extreme (G4) hunger index group has a significantly lower average per capita petroleum consumption rates than the moderate hunger index group (G2).

Here are the explicit results for the full 1990 data set:
Group
comparison
difference between means
SE (sqrt(ms/2((1/n1+1/n2))
q
crit q
(0.05,df,k)
Conclusion
G1 to G4
5.691775
0.595184
9.563057
3.685
reject
G1 to G3
5.469015
0.585667
9.338102
3.685
reject
G1 to G2
2.536409
0.641309
3.955053
3.685
reject
G2 to G4
3.155365
0.618819
5.09901
3.685
reject
G2 to G3
2.932606
0.609671
4.810142
3.685
reject
G3 to G4
0.222759
0.560951
0.39711
3.685
accept

The null hypothesis is rejected, for the all of the comparisons between groups, except for the comparison between the serious hunger index (G3) and alarming/extreme (G4) hunger index groups, which do not have a significantly different per capita consumption rate.

Excluding that strong outlier, Djiboti, did not change the conclusions for either the 2010 or 1990 data.

Here then is a summary of the average and standard deviations of per capita petroleum consumption rates as divided into the different hunger index groups (Djiboti included):
Average and standard deviations of per capita petroleum consumption (b/py)

GHI
1990
2010
G1: low
6.4±5.2
7.5±9.6
G2: moderate
3.8±2.9
4.8±3.1
G3: serious
0.92±0.79
1.1±0.9
G4: alarming/extreme
0.69±1.47
0.74±1.16

In conclusion, it is likely per capita petroleum consumption is significantly different between at least the low and moderate versus serious and extreme/alarming hunger groups.  Moreover, those countries with a global hunger index at the “serious” of higher level always have a per capita petroleum consumption rate of about 1 barrel per person per year of less.  
As a reminder, “serious hunger,” as defined by the IFPRI, means a GHI of between 10 and 20.  That, in turn, means that the weight average of: the percentage of the population that is undernourished, the percentage of underweight children under five and the percentage of children dying before the age of five, is in the 10 to 20% range.  That’s serious hunger alright. 

Unfortunately, the entire continent of Africa, right now, is just barely above an average per capita consumption rate of 1 barrel per person per year (see part 9 of my previous series) and the level of per capita consumption like to decline in the near future. This is at least in part due to the continued rate of population growth in Africa.  The need to step up food aid to Africa to avoid famine would appear to be right around the corner.  However, I am by no means convinced that this is going to happen in time to (see e.g., The Food Crisis, pointing out that promises of $22 billion over three year for food aid from the G20 countries is in serious shortfall).

Is the change in hunger index from 1990 to 2010 consistent with the expected change in per capita petroleum consumption?

The fact that I have pairs of hunger index data for a number of these countries allows me to answer another important question:  if the hunger index increased for a country from 1990 to 2010 (i.e., worsening hunger) did per capita petroleum consumption go down, and, if the hunger index decreased, did the per capita petroleum consumption go up?  

If the answer is "yes," then I would take this as further support of my contention that capita petroleum production is inversely related to food production—although this still doesn’t prove direct causation.

To test this hypothesis, I did a SIGNS test (to read more about an EXCEL implementation of the SIGNS test, see CHAPTER 15 Nonparametric Methods).

To set up the data properly, for the 99 countries for which we have data in both 1990 and 2010, if GHI increased and per capita consumption decreased, or, GHI decreased and per capita consumption increased, then I assigned a value of "1."  For the opposite scenario, if GHI increased and per capita consumption increased, or, if GHI decreased and per capita consumption decreased, 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 countiry are equal.  That is, if the proposed inverse relationship between GHI and per capita consumption is not true, then there should be an equal numbers of 1s and 0s (i.e., 44-45, in either group for a population of 99 countries).

My analysis showed that, of the 99 countries with paired data, 70 followed the proposed inverse relationship and 29 did not.  The lower and upper criterion values (i.e., a two-tailed SIGNS test) at a p < 0.01 are 37 and 63, respectively, and since 70 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 GHI with an inverse 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=55) with a GHI of greater than 10 (i.e., serious, alarming or extreme hunger) in 2010; the second subgroup was all countries (n=44) with a GHI of 10 or lower (i.e., moderate or low hunger).

Once again, the null hypothesis was rejected for either of these subgroups; at p<0.04 for the serious, alarming or extreme hunger subgroup, and, at p < 0.01 for the moderate or low hunger subgroup. 

This suggests that the inverse relationship between hunger index and per capit consumption holds for both low and higher level of hunger.

Summary and Conclusions

I found a statistically significant relationship between the hunger index and per capita petroleum consumption and this relationship holds for both the 1990 and 2010 data sets.  The relationship is an inverse relationship with hunger increasing when per capita petroleum consumption decreases, or, hunger decreasing when per capita petroleum consumption decreases. 

The change in hunger index and per capita petroleum consumption from 1990 to 2010 is also consistent with this inverse relationship, regardless of whether I consider countries with low to moderate hunger indexes or countries with higher hunger indexes.

A per capita consumption level of less than 1 barrel per person per year is related to a transition to serious or higher levels of hunger.  My speculation, is that the increased levels of hunger is related to the inability to adequately run the petroleum-driven food production system in the country.  But, I admit that other explanations may exist.  For instance, it is possible that increased hunger and decreased petroleum consumption may be both related to some third factor that is independently causing them both.  Some might think that decreased GDP would be such a third causative factor.  I happen to think that this is putting the cart before the horse—rather it is declining petroleum consumption that causes decreased GDP, but that’s a discussion best left for another time.

If per capita consumption falling below 1 barrel per person per year is an important trigger, or at least correlate of heighten hunger, then we can be both heartened and worried. 

Heartened, because per capita petroleum consumption in the USA, Canada, most of Europe, Japan and many other developed or developing countries, is presently much higher, sometimes more than an order of magnitude higher, than 1 b/py.  To me, that suggests that there is some slack in present consumption patterns, and therefore time to adapt to an alternative food production system, before the critical level of 1 b/py is reached and wide spread hunger occurs like in Africa now, in a post-peak oil world. 

Worried, because there are many countries, at least 40 from the International Food Policy Research’s survey of 2010, that already have a petroleum consumption level of less than 1 b/py.  Those 40 countries have a combined population of 2.3 billion, or about 1/3 of the world’s population.  Thirty-one of these countries are in Africa alone, although only three countries in the Asia-Pacific region— Bangladesh, India and Pakistan—account for 2/3 (1.5 billion) of this 2.3 billion combined total population. 

I expect that any diminution in food aid to these countries, or, reduced ability to produce their own food, due to the declining availability of petroleum to use in the food production system, will likely to push their populations farther into famine.