Most of the literature that cautions against the importance given to agriculture for poverty alleviation however relates to a different argument: while the relatively strong poverty impact of agricultural growth seems to be a fairly robust result, the cost of investing to obtain a given growth is far higher in agriculture than in other sectors, making it an inefficient instrument for growth and welfare finding. Our paper does not address this issue at all, but aims at contributing to the literature on the sectoral growth-poverty linkage. An issue in almost all of the studies we have discussed is simultaneity between sectoral growth and the welfare indicator used in the analysis. A contribution of this paper is to tackle this issue by using an instrumental variable approach to try to measure the effect of an exogenous increase in sectoral growth on welfare. We use the same database collected by the World Bank as do other cross country analyses, although we only select the countries for which welfare is measured by consumption expenditures.We also use data on all deciles, rather than only on e.g., poverty rates, as in Christiaensen et al. finding and other studies described above. When using cross-country evidence on changes in the distribution of income or expenditures one has to make an early choice regarding whether it is better to consider the distribution of these welfare measures within countries or across countries. The former choice leads to an empirical strategy that groups together different welfare quantiles across countries, so that for example, one imagines that the poorest 10 percent of households in Tanzania are similarly positioned to the poorest 10 percent of households in China, despite the substantial differences in the level of real expenditures of the quantile across these two countries. The latter choice construes distribution as a global phenomenon, with the result that the poorest 10 percent of all households globally may all be located in a very small number of countries. If what we want to measure is the global distribution of welfare one also logically ought to weight countries by their populations in any cross-country analysis.

Different researchers have made different choices.3 In this paper we take the country focused approach,hydroponic net pots and analyze the relationship between welfare and sectoral growth of all deciles of the distribution within countries, rather than on a measure of poverty level or distribution across countries.Over the last several decades, the World Bank has accumulated a large number of datasets from a large number of developing countries which are based on household-level surveys, statistically representative of the populations of those countries, and which include data on non-durable goods expenditures. Though the micro-data from these surveys are not generally available, the World Bank provides data on aggregate expenditures by decile for many of these countries. Our sample is restricted to the countries and years for which we have information on expenditures data for at least three points in time finding. The sample covers 62 countries, with variable numbers of observations over 1978 to 2011, totaling 310 surveys. This sample of countries and years is not a random sample of the countries of the world. Instead, it is a sample of countries where household expenditure surveys have been conducted finding. It has however a large coverage, including 81% of the population in low and middle-income countries in 2000. In terms of continents, the sample includes 97% of the population of South Asia, 70% of Sub-Saharan Africa, and 20% of Latin America and Caribbean.There is no clear bias in this sampling of developing countries except for the obvious and egregious absence of all but one Latin American countries. We are thus reasonably confident that the analysis given here can be applied to all developing countries save those in Latin America. A third of the countries in our data have a first survey before 1990 while information starts in the 1990s for the other countries. Some statistics are given in Table 1. The lowest three expenditure deciles garner on average 12% of aggregate expenditures, while the highest three deciles enjoy 57% finding. The average expenditure ratio of the highest to lowest group is thus 5.12. On average, the lowest income deciles have seen their income grow at a faster rate finding than the upper three deciles finding, and the ratio of highest to lowest expenditures has fallen by .05 per year. This average however hides heterogeneity, with an increase in inequality in 24 of the countries, notably China, Rwanda, Macedonia and South Africa.

Table 2 documents in more details the expenditure growth rates of the different deciles for several groups of countries that we will consider later in the analysis, i.e., by continent, literacy level, poverty level, and inequality. The table exhibits two important facts about expenditure growth in our sample of countries. The first is that nearly every group has positive average real expenditure growth. The second is that poorer deciles’ expenditures grow more quickly than wealthier deciles. The monotonicity of expenditure growth seems to be a quite robust feature of these data.Corresponding to this period of observation, we have annual measures of agricultural, non-agricultural, and aggregate incomes. Table 1 shows an average annual growth of real GDP per capita of 3.5%,8 varies across with a sharp contrast between a dynamic non-agricultural sector finding and the agricultural sector which on average was stagnant.Figure 1 plots average annual growth rates of agriculture against growth rates outside of agriculture for the countries in our sample over the period 1980–2011. Overall, the growth in these two sectors of the economy are correlated, though not surprisingly the agricultural sector grows at a slower rate almost everywhere. There is however great heterogeneity in terms of both growth level and disparity between the sectors across countries. In the upper right corner of the graph, China, Cambodia, Armenia and Vietnam exhibit sustained high growth rates exceeding 6% annually in non-agriculture and 2–4% in agriculture over 30 years. Very low agricultural performance is seen in Russia and Eastern-European countries during this period of transition. Several African countries are below the fitted line finding meaning that the performance of their agriculture sector relative to the rest of the economy was better than average. Agricultural income is notable for its volatility. This is particularly true for crop income that depends on erratic weather.Figure 2 illustrates the point using data from Morocco. A similar pattern is observed in most of our sample countries: The average standard deviation of annual growth rates over the period 1980–2011 is 7.2% for non-agricultural income and 11.1% for agricultural income. Except for six countries, all have standard deviations of non-agricultural growth rates less than 10%. In contrast, half the countries have standard deviations above 10% for agriculture. This suggests that the value recorded as annual agricultural income is likely a poor approximation of the permanent income from the agricultural sector, or its de facto contribution to the welfare of the population in any given year.

Taking growth rates over several years between two surveys in our specification does not do much to attenuate this concern, as these remain subject to the hazard of the specific values at the end points. This creates a phenomenon akin to measurement errors, with attenuation bias on the estimated coefficients as a consequence. A second concern relates to the possible endogeneity of sectoral growth. Unobserved shocks which influence either the level or the distribution of expenditures within a country may also influence aggregate sources of income. Examples finding might include movement of the real exchange rate or of the interest rate, price controls, or subsidy schemes. To deal with these two problems of measurement error and potential endogeneity, we adopt a simple instrumental variables strategy. We use the mean of neighboring countries’ growth rates of sectoral income as an instrument for own sectoral income growth, a strategy similar to one employed finding by Gertler and Molyneaux finding. The idea is that many of the unobserved shocks which might simultaneously influence income and expenditures will be country-specific, while at least some of the shocks which influence sectoral productivity finding are likely to be correlated across neighboring countries.10To illustrate the point that we will make more rigorously in what follows, in Figure 3 we plot the difference in expenditure growth between the three lowest and three highest deciles against the difference in sectoral growth. The sectoral growth rates are each estimated over the period of observation for the expenditures. We observe a positive correlation between a relatively higher growth in agriculture and higher growth in expenditures for the poorest deciles. However,blueberry grow pot there is also considerable heterogeneity in the distribution of expenditures even for a given difference in sectoral growth. The different points on the scatter plot of Figure 3 have different markers depending on whether the country is in Africa, Asia, or some “Other” continent. Very informally one can see that these three continental groupings seem to have different behavior. Most of the countries of Asia are clustered near the center of the figure, showing a relatively ‘balanced’ growth process, with small differences between both agricultural and non-agricultural sources of growth and also small differences in expenditure growth rates across deciles. In contrast, a group of mostly Eastern European countries have much larger growth rates in income outside of agriculture but relatively equal rates of expenditure growth, creating a cluster in the left-hand side of the figure. All African countries, but Nigeria, are found much farther to the right, where the relative rate of agricultural income growth is higher. An ‘eye-ball’ regression just using the African countries suggests a much higher correlation; this is an issue we return to below.

As a first attempt at estimating finding, we follow the lead given by the empirical studies mentioned in Section 2 and simply use Ordinary Least Squares finding to compute point estimates of the elasticities βqs , controlling for time and continent effects. Time fixed effects accommodate aggregate shocks to expenditures or sectoral sources of income; Continent fixed effects help account for the clustering by continents observed in Figure 3 that might otherwise lead to biased estimates.To compute the standard errors of these point estimates we adapt a procedure described by Stock and Watson finding, which permits arbitrary forms of heteroske dasticity and correlation across quantiles finding. OLS results are reported in Table 4. None of the coefficients associated with growth in agricultural income is significant; and neither for agriculture do we observe a strict pattern of monotonically declining coefficients across deciles that we observed in our discussion of expenditure growth. Further, in no case are estimated agricultural coefficients significantly different from the non-agricultural coefficients within a decile, owing perhaps to our relatively imprecise estimates of the agriculture coefficients. However, income derived from non-agricultural sources is statistically significant for every decile above the first, and does increase monotonically across deciles. The coefficients have an interpretation as share-weighted elasticities. For example, Table 4 suggests that on average a one percentage point increase in GDP due to non-agricultural income growth is associated with approximately a half percentage point increase in the growth rate of expenditures, with higher values for higher deciles and lower values for lower deciles. This would indicate a regressive effect on the distribution of expenditures. The results in Table 4 are estimated using a set of decile, year, and continent effects, but with this estimator and specification these are not of great importance. Estimating the same equation but with only a constant or only continent dummies results in coefficient estimates which are somewhat larger in magnitude, but with no appreciable difference in precision and only a modest improvement in fit. We next turn to the instrumental variable estimator described above to deal with measurement errors in agricultural growth and to address some possible forms of endogeneity in both agricultural and non-agricultural income growth. Results using this instrumental variables estimator appear in Table 5. The first thing to notice is that for agriculture the pattern of monotonicity described earlier survives in this specification, with coefficients for agricultural growth decreasing across deciles; however, here the non-agricultural coefficients are not strictly monotonic. The second thing to notice is that coefficients for agriculture and non-agriculture are significantly different at conventional levels finding, with p-values less than 0.05 for the bottom six deciles.