02 Sep The Rwandan Debacle: Disguising Poverty as an Economic Miracle
Recently the Financial Times published an investigation carried out by their data analysis team, which confirmed the findings that have been published on roape.net on poverty in Rwanda over several years. Of all the countries in the world for which there is data, only South Sudan has experienced a faster increase in poverty over the past decade. Rwanda’s official poverty statistics are verifiably false. The government, supported by the World Bank, is involved in a tragic debacle in which the poor are the real victims.
On 13 August 13 2019, the Financial Times published a lengthy investigation carried out by their data analysis team, which confirmed the findings that had been published on roape.net by several academics, regarding poverty in Rwanda. In particular, the Financial Times confirmed that the 7 percentage points decrease in poverty reported by the National Institute of Statistics of Rwanda (NISR) in 2016 , and endorsed by the World Bank in 2018, corresponded to an inflation rate of 4.71% for the period 2011-2014, that is, much lower than the total national CPI inflation for that same period (23%).
NISR responded by denying that they had deflated consumption by 4.71%, and saying that the comparison with CPI inflation was inappropriate because they had used a completely different methodology, called a Cost of Living Index (COLI), which ‘adjusts household consumption for each household, for each of the 12 months of the survey, in each of Rwanda’s five provinces.’ Surprisingly, however, they did not clarify by how much their COLI had deflated consumption between the two surveys in each of the five provinces, as would have been expected following their rejection of the FT’s figure.
NISR’s line of defense is almost identical to the one published a few months earlier by the World Bank, which had claimed, without providing alternative estimates nor explanations for the discrepancy, that ‘there is no clear theory to guarantee that the national average of COLIs and the national CPI need to be consistent’. In reality, a COLI is just a combined spatial and temporal price index calculated for the subsample of the population living in poverty, so there are, actually, very strong theoretical reasons why the two should be comparable and should, in most cases, yield fairly similar results, as pointed out by the World Bank’s former director of research and former acting chief economist, Martin Ravallion.
Despite these statements, most observers, including the authors of this blog, had, until recently, assumed this whole affair to be nothing more than a terrible mistake or an embarrassing oversight on part of the World Bank – or , as Justin Sandefur put it, a case of ‘NISR ma[king] convenient choices – perhaps choices WB couldn’t categorically call “wrong”.’ Any such doubts about the World Bank’s complicity, should now have been dispelled by their third public endorsement, without new supporting evidence, of NISR’s increasingly indefensible results. Indeed, in their latest press statement, published in response to the FT’s story, the World Bank did not only fail, for the third time, to address the legitimate and simple questions raised by various independent researchers, they decided to dig in their heels and accuse the FT of having used the wrong deflator:
the appropriate deflator for measuring poverty is not the Consumer Price Index or GDP deflator, but rather a composite “cost-of-living” index that is representative of the food and non-food consumption choices of poor households as well as the unit prices they face in the markets where they purchase goods and services. Poor households consume a diet that is less diverse and relies more on self-produced (especially in rural areas), basic, and cheaper staples.
This ambiguously worded press statement is particularly disingenuous and incriminating in that it hints, without stating so explicitly (because they know that that would be wrong), that the FT over-estimated poverty because they failed to account for the fact that poor people consume ‘cheaper’ goods than those included in the CPI. That is, in any case, how the statement was interpreted by Rwanda’s government officials, official media, and the army of Rwandan twitter trolls, who all quoted the World Bank press statement, claiming that it vindicated NISR’s numbers because the FT had used the ‘wrong deflator.’
As the World Bank’s experts know, however, what is at issue here are not the absolute levels of prices faced by poor households, but the rate of change in those prices. The fact that the World Bank did not, despite having all the data needed to do so, provide any indication as to what that rate of change might have been, is surprising, to say the least, since this could, by their own admission, have helped to settle this issue once and for all.
Fortunately, that evidence is already available in two blogposts on roape.net, including the work presented by Sam Desiere in 2017 and in another recent blogpost. This shows unequivocally that the (cheaper) food items consumed by poor people in rural areas increased more in price than the (expensive) items consumed by rich urban households, resulting in a food inflation of over 30% for poor households for the period 2011 to 2014. There is a simple explanation for this, which is that markets function better in urban areas, with better access to imported food, which makes them less vulnerable to adverse domestic production shocks.
And since poor households consume proportionally more food than non-poor households, that higher food-inflation rate weighed more heavily in the total final inflation faced by poor households, than in the inflation of non-poor households. This means that poor households faced unambiguously higher total (food + non-food) inflation than non-poor households over the period 2011 to 2014. This result is robust and holds for all publicly available price data sets (CPI, EICV, ESOKO), as the World Bank knows, since they have the data.
In other words, if NISR and the World Bank had computed their COLI correctly, it should have shown a total inflation rate for poor households that is higher than the national CPI inflation rate of 23% percent for this period, not lower as NISR and the World Bank suggest. The onus should therefore now be on the World Bank and NISR to explain why their estimate is lower than all alternative estimates of inflation that can be computed from publicly available data sources. They would also need to explain what price data source they used to compute their COLI, and why they think that this, as yet unnamed, price data source is more appropriate for estimating poor-inflation than the standard data sources that have always been used for this purpose in the past.
Below, we copy figure 1 from the World Bank (2018) paper, which plots the actual COLI used by NISR in each of the 5 Rwandan provinces over the 12 months of the EICV3 and EICV4 surveys (Integrated Household Living Conditions Survey). If NISR had accounted for inflation between the two surveys, their COLI should have jumped by an amount corresponding to the appropriate inflation rate in each province between the end of the EICV3 survey (October 2011, written as 1110 in the graph) and the start of the EICV4 survey (October 2013, denoted by 1310). As the graph clearly shows, however, this is not the case. In all provinces, except Kigali, the October 2013 COLI is almost identical to the October 2011 one. In the Southern province, it is even slightly lower, which would imply a negative inflation rate.
Source: World Bank (2018, p.17)
Now that these facts have been clarified, using the World Bank’s own graph, it would be appropriate for the World Bank to clearly state whether they still stand by their assertion that NISR’s COLI is an appropriate deflator? If so, they would need to explain whether they truly believe that poor rural households in Rwanda faced a negative to null inflation between October 2011 and October 2013 in all areas outside of Kigali? This would, of course, probably make Rwanda the only country in Africa, perhaps even in the world, to have experience virtually no inflation at all outside its capital for a full 2-year period.
Hiding behind the alleged “complexity” of poverty calculations, as the World Bank tried to do in their latest press statement, is simply not good enough for an institution that prides itself on the quality of its expertise. The issue under scrutiny (whether rural Rwanda experienced inflation or not) is actually fairly straightforward at this stage, and the magnitude of the disagreement is too large to be dismissed as trivial methodological quibbles, as the World Bank tried to do when it claimed that: ‘Differences over methodologies for poverty estimation are common in all countries, including developed ones’.
It is important to remember that, while the FT report only focused on the period 2011-2014, the estimates published on roape.net in 2019, which have not been challenged by any experts, have shown that poverty has continued to increase since 2014, and now stands at over 60% in Rwanda. This is, higher than when NISR started to measure poverty back in 2001 and up by as much as 15 percentage points since 2011. Of all the countries in the world for which there is data, we could only find one (South Sudan), that that has experienced a faster increase in poverty over the past decade. By contrast, NISR and the World Bank claim that poverty has decreased 8 percentage points over this period and now stands at 38%, which would make it a star performer comparable to some of the best Asian tigers. That’s a 23-percentage points difference between the two poverty estimates! If the World Bank’s experts are not even able to tell whether Rwanda is experiencing a catastrophic economic collapse like South Sudan, or an unprecedented economic miracle like the “Singapore of Africa”, then it could be argued that its expertise is of little or no value.
That is why the World Bank’s repeated insistence on publicly defending NISR’s obviously and verifiably flawed results, even when presented with mounting and robust adverse evidence, is so extraordinarily baffling and damaging. For there is something even more fundamental at stake here than the authoritarian model of development that Rwanda stands for. It is the status of facts and scientific evidence in public policy dialogue. In the age of fake news and alternative facts, the world needs, more than ever, to be able to rely on those institutions that are supposed to be the custodians of objective facts, which are the requisite common foundation for any viable democratic dialogue.
Whatever the final outcome of this saga might be, it is already clear that the reputational damage inflicted on the World Bank’s role as “leading provider of objective, unbiased data” will be immense, possibly irreparable, as had been feared by the group of World Bank whistleblowers calling themselves “Professionals for Truth in Aid”.
Unfortunately, it is not only the World Bank that is being less than transparent on Rwanda. For the past several years, the IMF has repeatedly exalted Rwanda’s GDP growth figures, despite failing to explain why growth in final household consumption reported in the National Account Statistics is so much higher than that reported in Household Survey data. In addition, bilateral donors have also been happy to hold up Rwanda as an African success story, despite mounting doubts about the veracity of the data underlying that narrative.
Ultimately, of course, the true victims of this debacle, are the Rwandan poor, whose plight over the past decade has been ignored by the very people who were supposed to help them, and passed off instead as an economic miracle, to satisfy distant imperatives of institutional expediency or political convenience.
The authors of this article have asked for anonymity.
Featured Photograph: A community meeting in Rwanda (6 March, 2018).
 For anyone wanting to verify these numbers, the COLI can be derived simply from NISR’s own datasets by looking at the difference between nominal and real consumption reported by NISR: COLI=cons1_ae/sol (where cons1_ae is nominal consumption per adult equivalent, and sol is real consumption in January 2014 prices). The values underlying the graph can be generated with the following command in STATA: table s0q19m id1, c(mean COLI), where id1 is the province variable and s0q19m represents the month of interview, which is contained in the household.dta file.
 The FT presumably chose to focus on this period because it was the only period for which a statistical manipulation could be easily proven, due to the discrepancy between NISR’s (2016, p.42) claimed rate of inflation of 14.0% (16.7 for food, weighted at .659, and 9% for non-food, weighted at .341) and the effective rate of inflation of 4.71% that could be deduced from the COLI (see World Bank, 2018, p.13, footnote 10). For the period 2014-2018, NISR did not indicate any inflation rate at all. This extremely unusual (even unheard of) move, means that it is not possible to show that they have misreported their own inflation rate for this period, since they reported none.