By Sam Desiere

In a recent blogpost an anonymous researcher on roape.net showed that poverty in Rwanda has increased from 2011 to 2014 by 5 percentage points. This contradicts the official poverty statistics and narrative, which claim that poverty decreased by 5.8 percentage points, namely from 44.9% in 2011 to 39.1% in 2014 (NISR, 2015). Importantly, the author published the Stata-files used to analyse the data of the EICV 3 and EICV 4 household surveys, enabling other researchers to verify his claims.

Recently, I also calculated trends in poverty using the same datasets. Although I used a slightly different (and, arguably, less sophisticated) methodology, the results confirm that poverty did not decrease. In addition, I show that the poverty trends are very sensitive to the inflation rate used. With an inflation of 16.7% (as reported by the National Institute of Statistics of Rwanda, NISR), poverty indeed decreased by at least 5 percentage points. With an inflation rate of 30% – which is in my view more in line with the ‘real’ inflation rate – my estimates show that poverty increased by 1.2 percentage points.

The fact that two researchers arrive – independently from each other – at the same conclusion, strengthens my belief that the EICV surveys show that poverty in Rwanda has increased. This has important implications for the current debate about (rural) policies in Rwanda, but I leave a discussion of these implications to researchers and policy makers more familiar with the reality on the ground and focus in this blogpost on the technical aspects of estimating poverty trends.

In this this post, I briefly describe my methodology and key findings and discuss (food) price inflation, which turns out to be a critical parameter. The Stata do-files required to replicate my findings can be found here.

Methodology

Rwanda’s poverty estimates are based on the Integrated Household Living Conditions Survey (EICV by their French acronym), which are conducted every three years. I used data from EICV 3, conducted in 2010/11 and EICV 4, conducted in 2013/14, which are made publicly available by the NISR. More specifically, I used the modules on food consumption purchased on the market and food consumption from own production. In both waves, the questionnaire of both modules is nearly identical. Food consumption is reported for more than 100 food items.

Unlike the anonymous researcher, I did not use the modules on non-food expenditure. I did so for two reasons. First, the NISR reports that most households spend over 60% of their budget on food. Hence, food expenditure is a good proxy of poverty. Second, non-food expenditure would require some additional data cleaning, which requires additional assumptions. Hence, I simply calculated food expenditure in both waves.

The meta-data of EICV 4 (available on NISR’s website) clearly stipulates that each sampled household in Kigali was visited 11 times over a period of 33 days. The modules on food consumption were administered during every visit. Rural households were visited 8 times over a period of 16 days. The meta-data of EICV 3, however, does not provide information on the number of times a household was visited. I simply assumed that the same methodology, for both rural and urban households, was followed in wave 3 as in wave 4. If this assumption is wrong – something I could not check – the results presented below will be erroneous.

In both waves, households reported how much they had spent on food purchased on the market by food item since the previous visit of the enumerator. I simply added up expenditure on all food items. Households also reported how much they had consumed from own production. Converting the consumption from own production in monetary values was more challenging. Households typically reported consumption from own production in kg. Some households also reported in the same module how much they would have paid on the market for this food item. I used this information to calculate the median, national price for each food item and used this price to convert consumption from own production in its monetary value. Since relatively few households reported prices, I did not attempt to calculate region specific prices nor did I correct for price seasonality. On this point my methodology differs from the anonymous researcher, who calculated a Laspeyres price index to account for spatial and temporal price variation.

To verify my assumptions, I checked whether my estimates of food expenditure are correlated with the household poverty status as reported by the NISR and included as a separate variable in the datasets. In both waves, food expenditure was lower for households classified by the NISR as extremely poor compared to household classified as poor, and the expenditure of this group was in turn lower than the expenditure of non-poor households. These results, available upon request, confirm that my assumptions are at least partially similar to the assumptions of the NISR.

Food expenditure can only be compared between the waves if the food inflation rate between 2010/11 and 2013/14 is known. I used two different inflation rates. First, I used an inflation rate of 16.7%, which is reported by the NISR (NISR, 2016, p. 43). Second, I estimated inflation based on food prices reported by the respondents, which I also used to convert food consumption from own production in monetary values. Inflation is then defined as a weighted average of the price increase of nine important crops. I used the same weights as those used by NISR to construct the 2013/14 adjusted food poverty line (NISR, 2015, table B4, p. 38). These estimates of inflation will be discussed in greater detail below.

Since I did not calculate total expenditure, but only food expenditure, I could not use the poverty lines proposed by the NISR. I therefore followed the ‘inverse’ methodology. First, I assumed that the NISR correctly estimated poverty in 2010/11 (44.9%) and used this information to determine the food expenditure threshold in 2010/11 prices that corresponds with this poverty rate. Second, I deflated food expenditure in 2013/14 using two different inflation rates, namely 16.7% and 30%. The first inflation rate corresponds with the inflation rate used by the NISR and thus allows me to replicate the findings of the NISR. The second inflation rate corresponds with my own estimate of inflation using the price data from EICV 3 and EICV 4. Third, I used the food expenditure threshold as an alternative to a poverty line to estimate the poverty rate in 2013/14. This approach is valid because I am not interested in ‘absolute’ poverty figures, but only in poverty trends.

In all analyses, I used the population weights to make the results nationally representative.

Results

Poverty trends

Using the EICV 3 and EICV 4 datasets, I calculated food expenditure per adult equivalent, respectively in 2010/11 prices and 2013/2014 prices. In order to estimate poverty trends, food expenditure in 2013/14 has to be deflated to express it 2010/11 prices. Poverty trends are very sensitive to the inflation rate used to deflate food expenditure. Results are presented for two inflation rates: (1) an inflation rate of 16.7% as reported by NISR and (2) an inflation rate of 30%, which is at the lower end of my inflation estimates based on ESOKO price data or EICV price data (see below for a discussion of inflation).

Figure 1 shows cumulative frequency distributions of food expenditure for these two situations, while table 1 summarizes poverty trends

With an inflation rate of 16.7% (left panel, figure 1), real food expenditure per adult equivalent increased for all households from 2010/11 to 2013/14 and, as a result, poverty decreased. Assuming a poverty rate of 44.9% in 2010/11 (which corresponds to a food poverty line of 100,232 RWF per adult equivalent), poverty decreased by 7.9 percentage points. This poverty reduction is even more pronounced than reported by official statistics, which states than poverty decreased by 5.8 percentage points.

With an inflation rate of 30% (right panel, figure 1), food expenditure does no longer increase between 2011 and 2014 for all households. Again assuming that poverty is 44.9% in 2010/11, poverty even increased by 1.2 percentage points.

Figure 1: Cumulative distribution of food expenditure per adult equivalent for EICV 3 and EICV 4 for an inflation rate of 16.7% and 30%

Table 1: Poverty trends in function of the inflation rate

In sum, the poverty trends are very sensitive to the inflation rate. With an inflation of 16.7% from 2011-2014, poverty decreased by at least 5 percentage points, which is in line with the official reports. With an inflation rate of 30%, poverty does not decrease. The question thus boils down to an accurate estimation of the inflation rate between 2011 and 2014.

Inflation rate

The EICV survey is not an ideal dataset to estimate inflation, because it does not contain much information on food prices. As explained earlier, some households report prices for those food items consumed from own production. This does not only mean that the number of observations is relatively limited, but also that households report prices of those items they did not buy on the market. I nevertheless used this information to calculate mean and median average prices by food item. I calculated national averages without taking into account price seasonality or regional price differences. In order to estimate ‘average’ inflation, a weighted average is taken over nine crops. The weights are proportional to the weights used for the construction of the 2013/14 adjusted food poverty line (NISR, 2015, table B4, p. 38). These nine crops account for 86% of the total calorific intake of the food basket. Two crops dominate this index: cassava (fermented) (weight: 38%) and dry beans (weight: 25%).

Figure 2 shows the increase in mean and median prices between 2010/11 and 2013/14 for nine crops, while the horizontal lines indicate the weighted average. The increase in median prices ranges from 10% for sorghum to 50% for cassava (both flour and roots). Median and mean inflation are 33% and 42%, respectively. This corresponds to an annual inflation of 9.5% and 12.5%, respectively.

Figure 2: Price increase for nine crops from 2010/11 to 3013/14 (mean and median prices)

These inflation estimates are substantially higher than the ones reported by NISR, which states that food prices increased by 16.7% between Jan 2011 and Jan 2014 (NISR, 2016, p. 43). Moreover, the estimates based on the EICV surveys are remarkably similar to the estimates based on detailed ESOKO price data, where I estimated inflation at 30.5% over the 2011-2014 period (details not reported here).

In sum, I believe that the ‘real’ food inflation rate is substantially higher than the one used by NISR to estimate poverty trends. This probably explains why I find that poverty increased, while the NISR reported that poverty decreased. These findings raise concerns, not only for Rwanda’s (rural) policies, but also for international donors that have presented Rwanda as a model for development because of the supposedly strong poverty reductions.

Sam Desiere is currently a senior researcher at HIVA, the research institute for work and society of the University of Leuven, Belgium. In 2015 he obtained a PhD in agricultural economics from Ghent University, Belgium, which focused on data quality of household surveys in developing countries.

Featured Photograph: As part of the DFID funded Vision 2020 Umurenge Programme (VUP), Rwanda’s flagship Social Protection Programme, women and men in northern Rwanda work on a public works site in 2012, building terraces to prevent soil erosion. 

Source: A Review of African Political Economy (ROAPE)

In his book entitled Poor Numbers, Morten Jerven cautioned against taking African development statics at face value, given the high political and financial stakes attached to these numbers, as well as the lack of institutional mechanisms to prevent political interference in many countries. Few countries illustrate his case more starkly than Rwanda. As An Ansoms et al pointed out in an article in the print issue of ROAPE earlier this year, ‘Statistics versus livelihoods: questioning Rwanda’s pathway out of poverty’ , the Rwandan government has used its record on poverty reduction and economic growth to legitimize its authoritarian rule and to deflect criticism of its human rights record, just as the previous regime had done up until 1990. Furthermore, Rwanda’s spectacular recovery after the genocide has made it somewhat of a “donor darling”, and has enabled the government to attract significant foreign resources in the form of aid from donors desperate to claim a share in this African success story.

Yet, questions have been mounting in recent years about the reality and sustainability of the “Rwandan miracle”, given the heavy-handed nature of the state-led agricultural transformation project (Dawson et al. 2016), and the government’s propensity for debt-financed investments in unproductive prestige projects, such as the Kigali Convention Centre. These questions came to a head in September 2015, when the National Institute of Statistics of Rwanda (NISR) published a poverty profile (NISR, 2015) based on the most recent household budget survey (EICV4 by its French acronym). The report claimed that the proportion of Rwandans living below the poverty line had fallen from 45% in 2010 to 39% in 2014, after a string of similarly successful decreases in the previous surveys. Two months later, Filip Reyntjens published a critique, claiming that the “decrease” in poverty had been artificially engineered by NISR by changing the type of poverty line used, from an “average” consumption basket based on actual consumption patterns of poor Rwandan households, to a “minimum” or “optimal” consumption basket, containing mostly highly caloric and inexpensive food types.

The change is not in itself problematic, as the choice of a poverty line is always, to some extent, arbitrary and there are many different acceptable ways to define a poverty line. The normative minimum consumption basket adopted by NISR is one such way. However, to make trend comparisons, all experts agree that it is crucial to use consistent methodologies, assumptions and definitions across time. Reyntjens claimed that had they done that, they would have found the proportion of people living below the minimum poverty line to have increased by 6 percentage points between 2010 and 2014. Unfortunately, Reyntjens never published the syntax files he used to compute his estimate. Neither did NISR accept to publish its own syntax files. Without this key piece of evidence, the debate has never been closed from a technical point of view, as it is impossible to show convincingly whether poverty has actually increased or decreased in Rwanda between 2010 and 2014.

We hope to contribute to settling this issue by publishing open, transparent and verifiable syntax files built using a publicly available dataset, which can be downloaded from NISR’s own microdata catalogue on its website (the two syntax files can be opened with .txt notepad or STATA software here). There are many ways to compute these things and there are innumerable adjustments and assumptions that must be made to arrive at an aggregate number. Consequently, it is difficult to replicate exactly the official estimates without access to the original syntax files. However, we hope that by submitting these to public scrutiny, such differences can be ironed out in an open and transparent manner, and any mistakes can be corrected to arrive at an estimate that all parties can accept. In constructing these estimates, our main priority has been to ensure consistency between the two surveys. We therefore try to use exactly the same code and assumptions in both years wherever possible. Below, we provide an overview of the key parameters and assumptions that entered the construction of these indices. Since there are several different poverty lines that have been generated by now, we decided to compute trends for all of them, namely:

• Average consumption basket: representing the minimum amount required to consume 2,500 kcal per day (adjusted for age and gender), using prevailing culinary habits of poor Rwandan households in 2001. This was the official poverty line used in 2001, 2005, 2010.
• Updated average basket: representing the minimum amount required to consume 2,500 kcal per day (adjusted for age and gender), using prevailing culinary habits of poor Rwandan households in 2014. This was the new poverty line computed by NISR in 2014, which should have been used in EICV4, but was never used because it was deemed too high.
• Minimum consumption basket: representing the minimum amount required to consume 2,500 kcal per day (adjusted for age and gender), using optimal (i.e. cheap and highly caloric) food types. This was the official poverty line used in 2014 (EICV4).
• Reyntjen’s poverty line: Reyntjens argued that since the minimum consumption basket was 19% lower than the updated average basket, trend comparisons with 2010 should have been made using a poverty line that was 19% lower than the one used in 2010.[1] For this poverty line, we did not construct a food basket, but simply calculated 81% of the figure from the total poverty line computed from the average consumption basket.

In all consumption baskets, the quantities and caloric values are kept constant across surveys. Prices for each item are given as the national median price across regions and across months, as reported in the auto-consumption module of the EICV survey (see table 3 below). Consumption aggregates have been adjusted for spatial and temporal price differences using a Laspeyres index (see table 2 below). The Laspeyres index was chosen because it yielded estimates that were closest to official poverty estimates in EICV3 for the average basket. The choice of price index does not affect the conclusions of this blogpost.

The results are reported in table 1 below. All poverty lines yield similar trends when used consistently over time, indicating that poverty increased between 5% and 7% points between 2010 and 2014. All changes are statistically significant at the 5% level.

It should be noted that our results differ from those obtained by simply updating the poverty line for inflation using CPI data, as was done by NISR in their 2016 trend report (NISR, 2016). In principle, if the data are of good quality and sufficiently disaggregated, both methods should be equivalent and should not yield significantly different results. This therefore raises questions about the quality / reliability of official CPI data, and/or the quality of price data collected by the EICV. In either case, this would undermine our ability to correctly estimate poverty levels in Rwanda. The discrepancies found here should invite us to more closely scrutinize official statistics coming out of the Rwandan statistical office. GDP growth figures appear to be incompatible with the findings of the EICV survey, given than agriculture still accounts for about one third of GDP and two thirds of the labour force.

Tables

Table 1: Summary of poverty lines and poverty rates

Table 2: Laspeyres price index by quarter and province (computed from price data in auto-consumption file)

Table 3: Food baskets used to compute poverty lines

The authors of this article have asked for anonymity.  

Featured Photograph: Parc National des Volcans, Rwanda. August 4, 2005

References

Reyntjens, F. 2015. “Lies, Damned Lies and Statistics: Poverty Reduction Rwandan-style and How the Aid Community Loves It.” Blog of 3 November 2015 posted on http://www.africanarguments.org.

NISR. 2015. Rwanda Poverty Profile Report 2013/2014: Results of Integrated Household Living Conditions Survey. Kigali: NISR.

An Ansoms, Esther Marijnen, Giuseppe Cioffo, and Jude Murison, “Statistics versus livelihoods: questioning Rwanda’s pathway out of poverty”, Review Of African Political EconomyVol. 44 , Iss. 151, 2017.

National Institute of Statistics of Rwanda (NISR), Poverty Trend Analysis Report, June 2016.

Jerven, Morten. Poor numbers: how we are misled by African development statistics and what to do about it. Cornell University Press, 2013.

Dawson, Neil, Adrian Martin, and Thomas Sikor. ‘Green revolution in sub-saharan Africa: Implications of imposed innovation for the wellbeing of rural smallholders.’ World Development 78 (2016): 204-218.

Notes

[1] Note that Reyntjens argument is not strictly speaking correct, since it would still require us to compare two different consumption baskets. To be methodologically sound, the 19% reduction would thus need to be applied to the same basket in both years, as we are doing here.

[2] In the average consumption basket, the non-food component is computed based on the average food share for households in the 7^th decile in 2001. In the updated and minimum baskets, the non-food components are computed based on the average food share for households in the 5^th decile in 2014.

[3] National median price of product as reported in the auto-consumption module.

Source: A Review of African Political Economy (ROAPE)