Accounting for the reduction in rural poverty in Ghana 1988-1992

84 * PSP Discussion Paper Series 19706 November 1995 Accounting for the Reduction in Rural Poverty in Ghana 1988-1992 Christine Jones Xiao Ye November 1995 Poverty and Social Policy Department Human Capital Development and Operations Policy The World Bank  Abstracts This Booklet of Abstracts contains short summaries of recent PSP Discussion Papers; copies of specific papers may be requested from Patricia G. Sanchez via All-in-One. The views expressed in the papers are those of the authors and do not necessarily represent the official policy of the Bank. Rather, the papers reflect work in progress. They are intended to make lessons emerging from the current work program available to operational staff quickly and easily, as well as to stimulate discussion and comment. They also serve as the building blocks for subsequent policy and best practice papers.  Preface This paper is one of a series undertaken as part of the Extended Poverty Study in Ghana. A synthesis of this work is available in Ghana: Poverty Past, Present and Future (Report No. 140504-GH, Population and Human Resources Division, West Central Africa Department, World Bank, Washington D.C., June 29, 1995). Four of the background papers are appearing in the Poverty and Social Policy Discussion Paper Series: Harold Coulombe and Andrew McKay, 'An Assessment of Trends in Poverty in Ghana, 1988-92.' Poverty and Social Policy Discussion Paper No. 81, World Bank, Washington D.C. (November 1995) Lionel Demery, Shiyan Chao, Rene Bernier and Kalpana Mehra, 'The Incidence of Social Spending in Ghana.' Poverty and Social Policy Discussion Paper No. 82, World Bank, Washington D.C. (November 1995) Andy Norton, David Korboe, Ellen Bortei-Dorku and D.K. Tony Dogbe, 'Poverty Assessment in Ghana using Qualitative and Participatory Research Methods.' Poverty and Social Policy Discussion Paper No. 83, World Bank, Washington D.C. (November 1995) Christine Jones, and Ye Xiao, 'Accounting for the Reduction in Rural Poverty im Ghana, 1988-1992.' Poverty and Social Policy Discussion Paper No. 84, World Bank, Washington D.C. (November 1995) T'he papers draw upon the work of many. The contributions of the Ghana Statistical Service, Ministry of Health, and Ministry of Education in the Government of Ghana, and of UNICEF and the Canadian International Development Agency (CIDA), are gratefully acknowledged. However, the views expressed are those of the authors. They should not be attributed to the World Bank, its Board of Directors, its management or any of its member countries. 2   Abstract Poverty appears to have declined in the three rural localities of Ghana between 1988 and 1992. It has been hypothesized that this was largely due to the growth in nonfarm self-employment in the context of a stagnating agricultural sector. This explanation is problematic, since there is no clear source of demand for the output from the rural nonfarm sector. An alternative hypothesis is that the food production did increase. This explanation also poses some problems, in view of the difficulty of accounting for the factors that would have led to an increase in production. These questions are the basis for the hypotheses explored in this paper. The first hypothesis is that the strong growth in per capita expenditure revealed by the GLSS data may be partially the result of changes in the design of the GLSS3 questionnaire which resulted in an upward bias in the GLSS3 data relative to GLSS 1/2 data. Thus the increase in expenditure is an artifact of changes in questionnaire design. The second hypothesis is that agricultural incomes, rather than stagnating, actually did increase, providing an impetus for the growth in rural nonfarm self-employment. The paper looks at altemative data sources on agricultural production and prices to come up with a view on what happened to agricultural revenues. The conclusion of the paper is that both hypotheses appear to have some merit. It is plausible that there was a small reduction in poverty due to the growth in agricultural incomes, but that the magnitude of the reduction is overstated by the GLSS data due to changes in the questionnaire that bias the GLSS3 estimates upwards relative to earlier years. 3   CONTENTS 1. Introduction 2. Questions About the Comparability of the Surveys 3. What Happened to Agricultural Revenues? 4. Accounting for the Very Large Reduction in Poverty in the Savannah 5. Conclusion Annex 4   1. INTRODUCTION Poverty appears to have declined in the three rural localities of Ghana between 1988 and 1992, according to the GLSS data (Table 1), as reported in the Extended Poverty Study (World Bank, 1995a). The Ghana Country Economic Memorandum (World Bank, 1995b) hypothesizes that the decline in poverty was largely a function of the growth in nonfarm self-employment in the context of a stagnating agricultural sector. This explanation raises some additional questions. If agricultural incomes did indeed stagnate, what was the source of demand for output from rural nonfarm self-employment? It would not appear to be the formal urban sector, since there was little growth in wages. The Extended Poverty Study raises the question of whether agricultural production did indeed stagnate. An increase in agricultural production would explain at least part of the growth in rural expenditures, but raises the question of what accounts for the growth in agricultural production. It is difficult to identify shifts in price incentives or nonprice factors that would have contributed to a strong increase in agricultural in the post 1987 period, since most of the large relative price changes took place during the initial years (1983-1986) of the Economic Recovery Program initiated by the government in 1983. So what does account for the decline in poverty? The first hypothesis explored in the paper is that the strong growth in per capita expenditure revealed by the GLSS data may be partly the result of changes in the design of the GLSS3 questionnaire which resulted in an upward bias (in estimated household expenditures) in the GLSS3 data relative to GLSS1/2 data. The GLSS data show a much stronger increase in expenditure, about 3.3 percent per annum between GLSS1 and 3, than the national accounts data, which show a growth rate of private consumption per capita of about 1.1 percent per annum between 1987 and 1992. Below we look at how the changes in the questionnaire design may have resulted in the apparent increase in expenditures. The second hypothesis is that agricultural incomes, rather than stagnating, actually did increase, providing an irnpetus for the growth in rural nonfarm self-employment. The paper looks at alternative data sources on agricultural production and prices to come up with a view on what happened to agricultural revenues. The conclusion of the paper is that both hypotheses appear to have some merit. It is plausible that there was a small reduction in poverty due to the apparent growth in agricultural food incomes, but that the magnitude of the reduction is overstated by the GLSS data due to changes in the questionnaire that bias the GLSS3 estimates upwards relative to the GLSS 1/2 data. It is worth noting that the changes in poverty in Ghana are largely due to the growth in mean expenditure, and not to an improvement in distribution of expenditure favoring the poor. This is evident from the decomposition of the change in poverty into its growth and redistribution components following the methodology used in Datt and Ravallion (1992), presented in Table 2. Between 1988 and 1992, the redistribution effect was adverse in the Rural Forest and Coastal localities, and in the Savannah, it was responsible for less than a quarter of the reduction in poverty. Similarly, the decline in expenditure accounts for most of the increase in poverty between 1988 and 1989. Only in the Rural 5  Table 1. Rural Ghana: Poverty Incidence and Poverty Gapa" ...................................................................... .............................................................I.......... ..................... ................................................................................................ Po Po Po PI PI Pi GLSSI GLSS2 GLSS3 GLSSI GLSS2 GLSS3 ...... ......................................................................................................... ............................................................................................................... Locality Rural Coastal 0.377 0.446 0.286 0.124 0.149 0.068 Rural Forest 0.381 0.419 0.330 0.114 0.128 0.083 Rural Savannah 0.494 0.548 0.383 0.178 0.204 0.105 Rural Ghana 0.419 0.467 0.339 0.138 0.157 0(087 a/ These estimates are based on adjustments for the effects of the shortened the recall period for food expenditures in GLSS3 (see World Bank, 1995a and discussion below). Coastal locality was there a large adverse redistribution effect that contributed to the increase in poverty. The Savannah region stands out as having the largest increase in poverty between 1988 and 1989 as well as the largest decrease between 1988 and 1992. Clearly any explanation for the change in poverty must also come to terms with the large change in welfare in the Savannah over this period. Table 2. Rural Ghana: Growth-Redistribution Decomposition Growth Distribution Total change Component Component Residual GLSSI-GLSS3 (1988-1992) Rural Coastal - 9.09 -6.67 0.65 -3.06 Rural Forest - 5.18 -8.38 4.65 -1.46 Rural Savannah -11.12 -8.39 -2.56 -0.16 Rural Ghana - 8.00 -7.70 1.21 -1.51 GLSSI-GLSS2 (1988-1989) Rural Coastal 6.90 5.13 2.86 -1.09 Rural Forest 3.78 2.11 0.32 1.36 Rural Savannah 5.37 8.54 -2.61 -0.57 Rural Ghana 4.85 4.93 0.43 -0.52 GLSS2-GLSS3 (1989-1992) Rural Coastal -15.99 -11.37 0.27 4.89 Rural Forest -8.96 -11.78 2.93 -0.12 Rural Savannah -16.49 -15.8 1.09 -1.78 Rural Ghana -12.85 -12.42 1.58 -2.01 6  2. QUESTIONS ABOUT THE COMPARABILITY OF THE SURVEYS Given the increase in rural poverty between GLSS1 and 2, the question arises as to which survey would make the better baseline for the purposes of comparison with GLSS3. One hypothesis that has been advanced to explain the decrease in expenditure between the first and second rounds of the GLSS is that there may have been sampling errors in GLS SI leading to a substantial undercount of the poor, thus biasing the GLS S 1 estimates upwards relative to the GLSS2 estimates. According to this hypothesis, GLSS2 would make the better baseline comparison for the GLSS3 data. However, the evidence on household size does not support the hypothesis that there was an undercount of poor rural households in GLSS1. Mean household size fell between GLSS I and 2 indicating that poverty estimates would tend to fall in GLSS2, all other things being equal (see Table 3). Moreover, food shares declined between GLSS1 and 2, also suggesting that poor households were not under-represented in the GLSS1 survey (see Table 4). The decline in the food share between 1988 and 1989 is unexpected, given the decline in total expenditures. Ordinarily, one would have expected the food share to rise as expenditure falls. The relative price of food and nonfood remained unchanged between 1988 and 1989, so that would not explain the declining food share. A possible explanation for the fall in expenditure between 1988 and 1989 could be that the food deflator was misestimated for the 1987/88 period. There is some evidence (discussed below in the section on producer prices) suggesting that the food deflator for the 1987/88 might not have captured the spike in staple food prices, particularly maize and cassava, two of the main staple food crops. If the food deflator were underestimated in 1987/88, the overall rural deflator would be biased downwards for GLSS1. This would bias expenditures upwards, thereby exaggerating decline in expenditure between 1988 and 1989. Moreover, had the food deflator been higher in 1987/88 than reported, the relative price of food to non food would have fallen between 1988 and 1989. A fall in relative food prices would also be consistent with the decline in the food share between GLSS1 and 2. If so, GLSS2 might be the better baseline survey. However, we do not have enough information about the construction of the food deflator and sufficiently disaggregated price data to reach a conclusion. As for the GLSS1/2 and GLSS3 comparisons, the evidence suggests that the change in expenditures may be biased upwards due to changes in the GLSS3 questionnaire design for which no corrections to the data were made. We look at two possible sources of bias: 1) changes in the recall period and 2) the more detailed breakdown of the expenditure data. It should be noted that the changes in GLSS3 may have improved the estimates of expenditure and thus in an absolute sense are less biased than estimates based on the GLSSI/2 questionnaire. Our concern in this paper is whether the changes in GLSS3 substantially affect the comparability of the data, by introducing a bias relative to the GLSS1/2 estimates, and if so, whether the estimates for GLSS3 are biased upwards or downwards relative to the GLSS 1/2. 7  Table 3. Rural Ghana: Mean Household Size, GLSS1-3 ..................................... ...... ........... ............................. -.-.... .............. ...... .................... ................... ................. .................... ........................................................................ Mean Household Size GLSS1 GLSS2 GLSS3 ................................................................... ................................................................... ......................................................................... ....................................................................................... Locality Rural Coastal 4.48 4.52 4.00 Rural Forest 4.86 4.36 4.38 Rural Savannah 5.80 5.67 5.46 Rural Ghana 5.05 4.75 4.60 Table 4. Rural Ghana: Per Capita Expenditure by Category, GLSS1-GLSS3 ............................................................................................................................................................................. ............................................................ Consumption of Purchased Total Food Non food Total Mear HH Share of Home Production* Food** Expenditure Expenditure Expenditure Food Expenditure ...........I........................................................ ..................................................................... ............................................................................................... Rural Coastal GLSSI 41265 88784 130049 59766 189815 69.3 GLSS2 29563 83518 113081 60897 173978 66.0 GLSS3 36815 98229 135044 82761 217805 63.0 GLSS2-GLSSl -11702 -15837 Rural Forest GLSS1 64068 68645 132713 53950 186663 70.8 GLSS2 42994 73535 116529 63713 180242 66.2 GLSS3 56074 72501 128575 84929 213504 61.7 GLSS2-GLSSl -21074 -6421 Rural Savannah GLSS1 87928 39752 127680 39406 167086 74.4 GLSS2 65343 41211 106554 38364 144918 74.0 GLSS3 72280 65856 138136 50333 188469 73.1 GLSS2-GLSS1 -22585 -22168 Rural Ghana GLSS1 67305 63126 130431 50246 180677 71.7 GLSS2 47224 65265 112489 54800 167289 68.7 GLSS3 57592 75653 133245 72526 205771 65.9 GLSS2-GLSS1 -20081 -13388 * Includes in-kind payment from employment. ** Food expenditure includes expenditure on alcohol, but not tobacco. Tobacco expenditures were subtracted from the aggregate food expenditure for GLSS3 used by Coulombe and Mckay (1995) to be consistent with the GLSS I and 2 aggregate. 8  Change in the recall period The GLSS3 questionnaire in general used a much shorter recall period in collecting expenditure data both for food and non food. For purchased food expenditures, the recall period was changed from 14 days in the GLSS1/2 questionnaire to 3 days. Based on work done by Scott and Amenuvegbe (1990) which showed that expenditures were a factor of 1.029 higher for each day the recall period was shortened up to a week, Coulombe and McKay (1995) multiplied the GLSS1/2 food purchase data by a factor of (1.029)5 to correct for the distortion introduced by the change in the recall period. Unlike food expenditures, however, the GLSSI/2 frequent nonfood expenditure data were not corrected for the shift in recall period from fourteen days in GLSS 1/2 to three days in GLSS3. If this correction were made, mean expenditures in GLSS1 would rise by 2,000 cedis (less than 1 percent), with a small impact on the incidence of poverty (see Table 5). For the sake of uniformity, this correction should also be made even though its impact on the poverty estimates is small. Another way that the change in recall period affected the estimates was the shift in certain items non food from the infrequent category of expenditure in the GLSSI/2 questionnaire to the frequent category in the GLSS3 questionnaire. We can get some idea from the GLSS1/2 surveys of the extent of the bias resulting from the shift from the infrequent to frequent section of the questionnaire. The surveys asked households to report their less frequent expenditures on an annual basis as well as what they actually spent in the last two weeks. Mean expenditure based on the two week estimates is about 60 percent higher than the annual estimate (see Table 6). It is sometimes argued that the annual estimates are preferable, since using the two week estimates introduces more variance and thus would tend to increase poverty, all other things being equal. While the coefficients of variation for the two week recall are about double those of the annual estimates, mean expenditures based on the two week recall are sufficiently higher such that the poverty incidence using the two week estimates differed by more than one percentage point in GLSS1 and 0.6 percentage points in GLSS2 from the poverty incidence based on the annual estimates. This only accounts for the bias in shifting from an annual to a two week recall period. A further bias is introduced in GLSS3, because the recall period used for frequent expenditures shifted from 14 to 3 days. Purchased food expenditures in GLSS1/2 were revised upwards by 15 percent to take account of the bias introduced in the change from a 14 day to a 3 day recall period. If the expenditures based on a two week recall are 60 percent higher than the estimates based on annual recall, and those based on a three day recall are 15 percent higher than those based on two week, then the combined effect of shifting nonfood expenditures from an infrequent (annual) to a frequent three day estimate would be an upward bias in the GLSS3 estimates of about 70 percent. 9  Table 5. Rural Ghana: Poverty Incidence Corrected for Change in Recall Period for Nonfood Daily Expenditure* Correctedfor Recall Error Not Correctedfor Recall Error Po Po Po P0 Po PO Locality GLSS1 GLSS2 GLSS3 GLSSI GLSS2 GLSS3 ................................. ....................................................................................... ........................ ................................... ............................................... Rural Coastal 0.368 0.434 0.286 0.377 0.446 0.286 Rural Forest 0.374 0.408 0.330 0.381 0.419 0.330 Rural Savannah 0.490 0.546 0.383 0.494 0.548 0.383 Rural Ghana 0.412 0.459 0.339 0.419 0.467 0.339 * The variable EXPDAY in file IE34 was multiplied by (1.029)5 for GLSS 1 and 2. Table 6. Rural Ghana: Per Capita Annual Other Expenditure, 2-week and 12-month Estimates Per Capita Infrequent Expenditure Poverty Incidence 2-week 12-month Ratio 2-week 12-month Estimates Estimates Estimates Estimates (a) (b) (a)/(b) Rural Coastal GLSS1 39010 25194 1.55 0.341 0.377 GLSS2 44468 23784 1.87 0.435 0.446 Rural Forest GLSS1 40814 24635 1.66 0.384 0.381 GLSS2 52749 29024 1.82 0.406 0.419 Rural Savannah GLSS1 20369 15745 1.29 0.481 0.494 GLSS2 21770 17963 1.21 0.555 0.548 Rural Ghana GLSS1 33472 21730 1.54 0.408 0.419 GLSS2 40755 24221 1.68 0.461 0.467 10  It is difficult to estimate what the overall impact of the shift from the infrequent to frequent categories is, given the lack of exact correspondence between the categories of expenditure used in GLSS1/2 and 3. However, two relatively important categories of expenditure, public transport expenditure and medical expenses (other than hospital- related medical expenses), shifted from the infrequent to the frequent section. Transport and medical expenditures ( 5,400 cedis) alone accounted for almost 3 percent of mean total expenditure in GLSSI, the bias could have amounted to about 15 percent of the 25,000 cedis difference in mean total expenditure between GLSSI and 3, assuming that the shift in recall period from annual to 3 day increases mean expenditure by 70 percent. And even for those items that remained in the infrequent category, the recall period in GLSS3 was reduced from one year to 3 months if the household purchased the item more than 4 times a year. This would presumably also introduce some bias in the estimate of infrequent expenditures, though we have no way to estimate the impact. A final way in which the recall period change affected the estimates was for consumption of home production. The method used to estimate consumption of home production (CBP) changed substantially between GLSSI/2 and GLSS3. The GLSS1 and 2 questionnaires asked for information concerning the number of months a home- produced good was usually consumed, the number of times per month it was consumed, and the average value of consumption each time the commodity was consumed. This information was then aggregated to arrive at an estimate of consumption of home production. The GLSS3 survey relied on a direct recall method in which households were interviewed every 3 days over a two week period regarding the value of home production consumed since the last interview. These responses were then multiplied by 26 to annualize the data. The latter procedure introduces greater variance in the estimates of CHP, to the extent that some households may be interviewed in a period in which they are not consuming a home-produced commodity that they consume during other months of a year. Its advantage is that it offers greater precision in the estimates of what was actually consumed. There are various ways in which the comparability of the GLSS1/2 and GLSS3 estimates could be affected. Errors in estimating consumption of home production could be introduced in the GLSSI/2 method of estimation if a household does not correctly estimate the number of months it consumed a home-produced good, the number of times per month, or the average value of consumption each time a commodity was consumed. While the evidence suggests that lengthening the recall period tends to reduce the estimate of consumption, it is not entirely clear in this case what the impact would be of going to the three day recall period used in GLSS3 since the method of estimating home consumption in GLSS1/2 is the product of three values. If households systematically underreport the number of months they consume, but compensate by overestimating the number of times per month they consume or their consumption each time, then the resulting differences between the two methods of estimation would be smaller. 11  We can get some idea of the bias introduced by the difference between the reported months of consumption and the actual months a good is consumed from the GLSS3 questionnaire. For each home-produced item, the GLSS 3 questionnaire first asked households whether they consumed of the item in the last 12 months, and then asked a second question as to whether they eat the item all through the year or only in some months, and if so, which months. While the question on the number of months is not phrased specifically with respect to the last twelve months, the fact that it follows the question on whether the household consumed any of the home-produced item in the last twelve months introduces the possibility that the household will reply with respect to the last twelve months. If we assume that this is the case, then under the assumption that households are interviewed evenly throughout the year, the percentage of households with non-zero home consumption of a given item multiplied by twelve months provides an estimate of the mean "effective" months of home consumption.' This can then be compared with the mean reported months of consumption, which are presented in Table 7. We also computed the mean "effective" and reported months after correction for the seasonality bias introduced by non-uniform sampling throughout the year. Results are reported in Table 7. In general the differences between the seasonally unweighted and weighted ratios are very small. 12  Table 7. Rural Ghana: A Comparison of Effective and Reported Months, 1992 Not Adjustedfor Seasonality Adjustedfor Seasonalty Effective Actual Ratio of Effective Actual Ratio of Months Months Eff/Actual Months Months EfifActual ........... ....................................................................................................................................................................................................... Rirtal Ghana: nce 0.91 0.6 1.51 0.84 0.6 1.39 maize 5.02 4.4 1.14 5.22 4.4 1.19 millet 2.11 1.3 1.62 2.23 1.4 [.59 cassava 7.61 6.7 1.14 7.46 6.6 1.13 yarn 3.52 2.9 1.22 3.39 2.8 1.21 cocoyam 4.02 3.5 1.15 3.93 3.5 1.12 plaintain 4.63 4.0 1.16 4.61 4.0 1.15 Rural Savannah: rice 2.12 1.3 1.63 1.90 1.3 1.46 maize 5.97 4.3 1.39 6.20 4.4 1.41 niillet 6.09 3.7 1.65 6.46 4.1 1.58 caasava 4.66 4 1.17 4.03 3.5 1.15 yam 4.63 3.2 1.45 4.10 2.9 1.41 cocoyam 1.40 1.2 1.16 1.10 1 1.10 plaintain 1.34 1.2 1.11 1.11 1 1.11 Rural Forest: rice 0.39 0.4 0.98 0.41 0.4 1.02 maize 4.52 4.6 0.98 4.71 4.6 1.02 niillet 0.00 0 N/A 0.00 0 N/A cassava 10.01 8.9 1.12 10.09 8.9 1.13 yamn 3.91 3.5 1.12 3.85 3.6 1.07 cocoyam 7.09 6.3 1.13 7.04 6.3 1.12 plaintain 8.03 7.1 1.13 7.92 7 1.13 Rural Coastal: rice 0.00 0 N/A 0.00 0 N/A maize 4.50 4 1.12 4.69 3.9 1.20 millet 0.00 0 N/A 0.00 0 N/A cassava 7.39 6.6 1.12 7.56 6.6 1.15 yamn 0.91 0.8 1.14 1.28 1 1.28 cocoyam 1.89 1.5 1.26 2.01 1.5 1.34 plaintain 2.89 2.3 1.26 3.41 2.5 1.36 Source: GLSS3 The mean effective months of consumption were about 15 percent greater than the mean reported months for most of the major staple food crops. For millet/sorghum, the difference is about 60 percent. Thus, the procedure used to calculate the home consumption would tend to bias the mean GLSS3 subsistence consumption upwards relative to the method employed in GLSS1/2, assuming that households were reporting the number of months they consumed an item over the last twelve months (and assuming there was no offsetting bias in the GLSS1/2 estimates in the direction of an overestimate 13  of the number of times per rnonth or the value of consumption each time the item is consumed). If we adjust mean subsistence expenditures for the major staple food crops to reflect the proportionate difference between reported and effective consumption, nnean subsistence expenditures would fall by about 7,400 cedis (see Table 8a). The difference in mean total expenditures between GLSS1 and GLSS3 would be reduced by about 30 percent, and between GLSS2 and GLSS3 to almost 20 percent. The difference is particularly large for the Savannah: correcting for it would reduce the difference in mean expenditures between GLSS I and 3 by more than 50 percent (see below). Table 8a: Rural Ghana: Per Capita Staple Food Consumption Revised for Reported Month Bias All Rural Areas Home consumption of self-produced food .............................................................................................................................................................................................................................. Millet! Coco- Total Rice Maize Sorghum Cassava Yams yams Plantain Stcple ........ ................................................................................................................................................................................................................... GLSS1 732 10879 9203 14761 5248 4764 6452 52039 GLSS2 722 7507 6900 9308 4198 3401 3923 35959 GLSS3 1102 4538 5069 11518 6731 4424 6632 40014 GLSS3 Revised 739 3834 3073 10210 5069 3860 5803 32588 GLSS3 Revised - GLSS3 -363 -704 -1996 -1308 -1662 -564 -829 -7426 To obtain an idea of how important a source of bias this might be, we adjusted the home-produced consumption of the main staple foods for each households by the ratio of mean reported to effective months of consumption. The poverty estimates were then recalculated (see Table gb). The poverty index for rural areas for GLSS3 rises by 2.7 percentage points, implying a reduction of 5.3 percentage points between GLSS1 and 3 instead of 8 percentage points. The Savannah showed a major change: poverty was reduced by only 6 percentage points, instead of 11 percentage points. This suggests that the change in the questionnaire with respect to the calculation of the consumption of home production could be a relatively important source of bias. 14  Table 8b: Rural Ghana: Poverty Incidence With Per Capita Staple Food Consumption Revised and Unrevised for Reported Month Bias Revised Unrevised Po Po Po Po Po Po GLSS1 GLSS2 GLSS3 GLSS1 GLSS2 GLSS3 .............I.......................................................................I........................... ............... I...............................................I.......................................... ..... Rural Coastal 0.377 0.446 0.294 0.377 0.446 0.286 Rural Forest 0.381 0.419 0.347 0.381 0.419 0.330 Rural Savannah 0.494 0.548 0.436 0.494 0.548 0.383 Rural Ghana 0.419 0.467 0.366 0.419 0.467 0.339 Of course, this is just an indicative calculation, as there remain many unknowns, including the question of how households actually interpreted the question on the number of months they consumed a home-produced item. It would seem, however, that there are enough questions about the comparability of the methods to raise serious question about the comparability of the data on consumption of home production. Expanding the number of categories A second large change in the questionnaire between GLSS1/2 and 3 was the large expansion in the number of items listed in the questionnaire. The impact of this expansion on expenditure has not been assessed. A study carried out in Indonesia suggests that it may have little impact, while a study on El Salvador came to the opposite conclusion. For some of the categories in the GLSS3, an increase in the number of expenditure subcomponents may not have mattered too much, since mean consumption for some of the subcomponents was very low. However, there are a few categories in the food expenditure section, for example, that might have been affected by the increase in the number of categories, as the subcomponents turned out to be fairly important in their own right. GLSS1 and 2 asked one question about consumption of alcohol beverages. The GLSS 3 questionnaire broke it down into six categories. Between 1988 and 1992, real mean per capita consumption of alcoholic beverages went up from 2925 to 5296 cedis, an 81 percent increase (see Table 9). Another example is dry pepper. There was no specific item in the GLSS1/2 question on consumption of dry pepper, so presumably households would have reported it under "other food expenditure." It was listed as a separate item in the GLSS3 questionnaire, however, with mean consumption of 995 cedis, a rather large food expenditure to relegate to "other," which amounted to only 196 cedis in GLSS1. A third example is the consumption of non dairy oils/fats. There were two questions on consumption of refined oil/fats in GLSS1/2 (palm oil and shea butter; refined oils) compared with seven questions in GLSS3 (coconut oil, groundnut oil, palm kemel oil, red palm oil, shea butter, animal fat, other vegetable oils). Purchases of oil/fats between GLSS 1 and GLSS3 in real terms went up from 2560 to 4039 cedis. Additional expenditures on these categories of expenditure alone represent about 20 percent of the increase in mean per capita food expenditure between GLSS1 2 and 3 . 2 It is possible that the price of these items may have increased substantially more than the overall CPI, explaining part of the apparent increase in consumption. While this seems unlikely, it is worth investigating. 15  Table 9. Rural Ghana: Per Capita Food Consumption by Categories .................................I.... ................................... .......I...................................................... ................................................................I.......................O...............I.................... . . . . . . . I. .. . . .. . . . . .. . . .. . . . . . . .. . . . .. . . Pulse Oilseeds Oi/ Fruits Vege- Meat Poul-try Milk Beverage Misc Alcohol Cereal Starch Fish Sugar Spice Prepared TOT1 Fats tables food fi m..e.consump tion..ofse"fp.. odu"edfood * ........................................................................................................................................................................................................................................................... Home consumption of self-produced food s GL.SSI 1974 1257 630 1998 5334 987 1160 64 29 227 223 20814 31305 812 66814 GLSS2 1476 766 483 1389 4106 836 1079 34 1 72 132 15129 20929 242 46674 GLSS3 3684 1640 541 1199 5374 1335 2665 49 8 168 10710 29454 581 57408 Consumiption ofpurchasedfood GLSSI 1872 1182 1930 709 6983 3374 1161 925 285 196 2702 8464 8041 15749 2044 1219 6289 63127 GLSS2 1687 1039 2145 692 7105 2987 1194 849 343 457 2862 7655 7488 18798 2033 1201 6733 65267 GLSS3 2195 1860 3498 593 6404 3810 1761 812 1115 250 5128 11856 8184 17942 1610 2402 6233 75653 Total consumption GLSSI 3846 2439 2560 2707 12317 4361 2321 989 314 423 2925 29278 39346 16561 2044 1219 6289 129941 GLSS2 3163 1805 2628 2081 11211 3823 2273 883 344 529 2994 22784 28417 19040 2033 1201 6733 111941 GLSS3 5879 3500 4039 1792 11778 5145 4426 861 1123 250 5296 22566 37638 18523 1610 2402 6233 133061 % Change GLSSI-3 53% 43% 58% -34% -4% 18% 91% -13% 258% -41% 81% -23% -4% 12% -21% 97% -1% 2% * Excludes in-kind payment for employment. 16  The number of categories in nonfood expenditures also increased substantially. There were 45 separate categories of frequent expenditure in GLSS3, compared with 9 in GLSS1/2. There were 63 categories of infrequent expenditure in GLSS3 compared with 23 in GLSS1/2.3 To examine whether there are any large increases that appear to be out of line, the expenditure estimates need to be redeflated by the price index relevant to a particular subcompenent since a large increase in consumption may be partly due to a large price increase relative to the overall increase in the CPI. Table 10 breaks down nonfood expenditure by category and also shows redeflated expenditures in GLSS1 and 2 by the relevant subcomponent of the CPI index to give a more accurate picture of the increase in each subcomponent. Medical expenditures increased the most -- 135%, but how much due to the shortening in the recall period and increase in the number of questions unclear. Table 10. Rural Ghana: Real Per Capita Nonfood Expenditure by Categories ............................................................................................................................................................................................................................................................ .H. . ...... ... ... ... ... .. ......H.............TFTA......... Cloting ..... 11.Gods ... Transp0zion ... aicaL ..... Recrt ...L Edcaon ... MivceJ1aneous.. T ba-co .. T.TAL AlRural GLSS1 145 3624 9626 7383 2617 2799 1793 2733 5080 2386 38186 GLSS2 218 3450 11248 7355 2582 2830 2200 3330 5752 2148 41113 GLSS3 764 6832 13301 9181 7300 6635 3253 5574 6458 1468 60766 % change, GLSSI-3 427% 89% 38% 24% 179% 137% 81% 104% 27% -38% 59% Redeflated GLSS3 16942 11878 4785 6583 2489 4265 7870 % change, GLSS1-3 76% 61% 83% 135% 39% 56% 55% Rural Savannah GLSSI 65 3020 7402 5256 1873 1884 1216 1616 4349 3161 29842 GLSS2 55 2547 7967 5264 1405 1797 1031 1645 5664 2946 30321 GLSS3 271 6152 9075 6940 4222 4596 2360 2740 4729 1810 42895 % change, GLSS1-3 317% 104% 23% 32% 125% 144% 94% 70% 9% -43% 44% Redeflated GLSS3 11559 8979 2767 4560 1806 2097 5763 Rural Coastal GLSSI 151 5033 10212 8828 3745 3057 1911 3795 6055 1800 44587 GLSS2 607 4993 10961 8500 2979 2720 2447 3819 5495 1802 44323 GLSS3 1665 8331 12343 10778 7824 7633 3755 8387 6808 1331 68855 % change GLSSI-3 1003% 66% 21% 22% 109% 150% 96% 121% 12% -26% 54% Redeflated GLSS3 15722 13944 5128 7573 2873 6418 8297 Rural Forest GLSSI 204 3407 11043 8312 2644 3374 2177 3075 5171 2076 41483 GLSS2 139 3323 13796 8300 3240 3643 2930 4313 5947 1742 47373 GLSS3 717 6646 17051 10163 9448 7747 3709 6439 7639 1267 70826 % change, GLSSI-3 251% 95% 54% 22% 257% 130% 70% 109% 48% -39% 71% Redeflated GLSS3 21719 13148 6193 7686 2838 4927 9309 Note: Redeflation factor is the change in the CPI between 1988-92 divided by the change in the subcomponent CPI over the same period. 3 Three items i' the infrequent expenditure section of the GLSSI/2 (weddings, funerals, taxes) were shifted to another part of the GLSS3 questionnaire and four other items were not included in the total, bringing the total number of items in the infrequent section of the GLSSI/2 questionnaire for purposes of comparison with GLSS3 down from 30 to 23. 17  3. WHAT HAPPENED TO AGRICULTURAL REVENUES? Thus, it would seem that while there may have been some growth in expenditures between GLSS 1/2 and GLSS3, there is reason to think that the rate is probably not nearly as large as data suggest. One approach to checking whether the growth in expenditure is reasonable is to examine what happened to the main sources of agricultural income. If agricultural incomes did indeed stagnate, it is unlikely there was much growth in rural incomes. On the other hand, if agricultural incomes showed a strong increase, one would also expect to see an increase in household expenditures. A problem with relying solely on the agricultural income data in the GLSS survey is that it is heavily influenced by the estimation of consumption of home production. If there are biases in that data, then agricultural income will be sirnilarly biased. So we looked at other sources of agricultural production and price data to investigate what happened to agricultural revenues. Production Trends Unfortunately, the analysis of what happened to agricultural production over the 1.987/92 period is complicated by the fact that there exist two series on agricultural production: one published by FAO and the other published by the Ghana Statistical Service (Appendix table 1). For most crops, they are quite sirnilar in the second half of the 1980s. The two series differ principally in the estimates for rnillet, sorghum and maize production in 1988 (see Charts 1,2,3). With respect to maize the FAO estimate is 25 percent higher than the GSS estimate, whereas the FAO estimates are lower for millet and sorghum, and dramatically so for millet. The discrepancies are more important in the 1990s, however. According to the GSS data there is a large jump in yam and cassava production in 1991-93 relative to the 1987-89 levels, whereas the FAO estimates show a decline for yam and a more modest increase for cassava (see Charts 4 & 5). According to the GSS data, yam production increased by 119 percent between 1991 and 1989 while cassava increased by 72 percent. Moreover, this was not a one year increase: production levels for both yam and cassava, according to the GSS data, remained at these higher levels through 1994. Both sets of data also show a considerable increase in maize production: in 1991, 1993, and 1994 production exceed 930,000 thousand metric tons compared to production in the range of 600,000 to 700,000 thousand metric tons in the 1987-1989 period, an increase of 33 - 50 percent. Both sets of data indicate that production dropped in 1990, a year of poor rainfall. For plantains, there is no difference in the two series after 1985 (see Chart 6). 18  Millet Production Sorghum Production 200000 350000 180000 - 300000 - 140000 - ..... . . .. .... 250000 ... .... ........ .. .......... . - 120000 .... . .. .. . 100000 ... ......200000 - 80000 .. - -- - ............. 150000 ...... .../ .. 6 0 0 0 0 .-- .... .\ 1 ------ --- -- -............................. ..... . .......... ....... 40000 , 100000 ............ . ....... ... ... .......... ....... 1980 1982 1984 1986 1988 1990 1992 1994 50000 Li 1980 1982 1984 1986 1988 1990 1992 | FAO - GSS| | FAO . ES|] Chart 1 Chart 2 Maize Production Plantain Production 1000000 1800000 800000 - ----- ---------- -- --- . . --- 1600000 *--- --.--.----- 800000 ..... ............ .. ..................................... ......140000........... ........ ..0...... .... .. ........ . ....... 19 /> / 1400000 *-.- . - 600000 -------- -- - --;--- ----IA 600000 . ........ ../ .............. ...... 1200000.....'.. 400000ooooo - . .... . . .. . . . 200000 ... . ........................ 800000............. . 0 600000 I I I I I I I- - 1980 1982 1984 1986 1988 1990 1992 1980 1982 1984 1986 1988 1990 1992 | _ FAO - GSS I |-FAO GSGSS] Chart 3 Chart 4 Cassava Production Yam Production 6000000 3000000 . 1 ~~2500000 . . .- 5000000-. .. ... I ~~~~~~~~~~~~~~~~~~~~2 5 0 0 0 0 00 . ....... .............. ..... ......... ................ ..... ......... . ............ 4 0 0 0 0 0 - ---------2 0 0 0 0 00------............ 4000000 - .. _.. 1500000 . ........ ... ................ ..... ......... . . 3000000 - - 1000000 . . . ....---- -- - 2000000 . .................... ........ 500000 . ...................... 1000000 O 1980 1982 1984 1986 1988 1990 1992 1980 1982 984 1C86A1R 1 390 1992 FAO GSS |FAO GS Chart 6 Chart 5 19  Producer price trends Producer price trends may help to shed some light on which production series is likely more accurate. The price data appear to be consistent with an increase in staple food production after 1990. They do not however provide strong support for the view that there was a spectacular increase in yam production such as is suggested by the GSS data. Table 11 and Chart 7 show producer prices for the four major crops, deflated by the rural CPI index.4 For maize and cassava, real producer prices peaked in 1987, following the drought in 1986, then came down between 1988-89, rose again in 1990 due to the drought and then fell in 1991. This is consistent with the pattern of production. Between 1987 and 1989, maize and cassava production increased by roughly 21 percent for both crops, and prices fell by roughly 50 percent. Real maize and cassava prices increased with the drought in 1990, and then fall again in 1991. The 1991 prices are roughly similar to the 1989 prices. To the extent that both cassava and maize are among the most price inelastic comnrnodities, one would expected a further fall in prices in 1991 and 1992 if there had been another major increase in output. The price data would seem to indicate that there has not been a dramatic increase in production of either crop. Table 11. Rural Ghana: Producer Prices Deflated by Consumer Price Index ........................... ...... .......................................... I..................................... ..................................................................................................... Year Plantain Yam Cassava Maize 198 107 92 76 90 1986 108 114 142 114 1987 138 138 215 153 1988 130 129 126 118 1989 137 146 105 82 1990 155 123 142 129 1991 105 110 111 89 1992 100 100 100 100 As for yams, real yam prices peaked in 1989, and then began to drop in 1990, and continued to fall in the 1991 and 1992. Between 1989 and 1992, yam prices fell by 53 percent. It is somewhat surprisinp that prices were not higher in 1990, given the decline in production in 1990 due to the drought. Since yam production increased dramatically in 1991 according to the GSS data, it is somewhat puzzling that the decline in yam price began in 1989. The production and price data do not seem totally consistent, but the large decline in real yam prices suggests that there has been an increase in yam production. 4 Producer price data are taken from Leenhardt (1993). They are annual data. Ideally one would use monthly data and look at price trends more closely ied to the crop calendar. 20  Real plantain prices rise in 1990, which is consistent with the poor harvest reported in 1990. Prices then fall sharply between 1990 and 1991 as production recovered to the 1988 level. Given the strong decline after 1990, however, in plantain prices, one would have expected to see an increase in plantain production in the 1990s compared to the second half of the 1980s. Real Producer Prices 240 220 ................... .. 200 . A.. . 160 12 0 -; : ..... .... ..... ........ . o s 140 ...... 120 100 80 1985 1986 1987 1988 1989 1990 1991 1992 -Pantain -. Yam - cassa _ Chart 7 Producer Revenues Using the GSS production data, we find that producer revenues fell for three out of the four main staple food crops 1987 and 1988 (except for plantains) (see Table 12). There was a further drop between 1988 and 1989 for cassava and maize. Weighted by each crop's share in total rural consumption, crop revenue from these four crops dropped by more than 10 percent between 1988 and 1989--due, except for plantains, to a fall in real crop prices. Revenues rebounded between 1989 and 1992, as the increases in production outweighed the drop in prices. Producer revenues increased by nine percent between 1988 and 1992, and by twenty percent between 1989 and 1992. The FAO data, on the other hand, indicate that income from these staple crops fell by 20 percent between 1988 and 1989 and by three percent between 1989 and 1992. 21  Table 12. Ghana: Real Producer Revenue GSS Production Data FA 0 Production Data Index of Real Crop Revenue Weighted Index of Real Crop Revenue Weighted Year Plantain Yam Cassava Maize Total* Plantain Yam Cassava Maize Total* --i... ... ........... .............-. ........ ............. ...-........ ........................ ............... ........-.. ..... ... ... ... ........... ................ ................ ............................................................. 1988 145 66 73 97 91 145 154 104 121 122 1989 131 75 62 81 79 131 187 88 81 103 1992 100 100 100 100 100 100 100 100 100 100 * Weighted by follow ing consumption shares of respective crops: ................................................................. .......................... ................ I................... Year Plantan Yarn C'assava Maize .... ...................................... ......................................................................................I........ 1988 016 0.14 0.41 0.29 1989 0.13 0.12 0.39 0.35 1992 0.19 0.21 0.40 0.20 Given the likely possibility that there was some increase in yam production in 1991, there is reason to think that agricultural incomes from the staple food crops may have increased between 1989 and 1992-- probably not as much as the 20 percent indicated by the GSS production data, but probably more than indicated by the 3 per cent decline based on the FAO data. According to the GLSS data, the increase in agricultural incomes between 1989 and 1992 was 18 percent (see Table 13). Remember, however, that the consumption of home production estimates for GLSS3 may be biased upwards relative to GLSS 1/2. Since consumption of home production accounts for a large share of agriculture income, if such a bias exists, then the recovery in agricultural incomes in 1992 would be less than what is indicated by the GLSS data. Whether or not agricultural incomes increased between 1988 and 1992 is more questionable. The GSS producer revenue estimate shows a 10 percent increase, while the FAG estimate gives a 25 percent decrease in income between 1988 and 1992. According to the GLSS data, agricultural incomes fell by 22 percent. However, there is an anomaly in the GLSS agricultural income for the rural coastal region; consumption of home production was an unusually low fraction of agricultural income for the rural coastal region for 1988, suggesting that agricultural incomes from sources other than consumption of home production might have been upwardly biased in 1988. If so, this would tend to overstate the fall in incomes between 1988 and 1992. Another factor that would tend to overstate the fall in incomes would be a bias in the CPI deflator for GLSS1. If the CPI deflator were underestimated for the GLSSI, agricultural incomes would also be biased upwards in GLSS1, thus exaggerating the decline in income between 1988 and 1992. However, shifts in the method of estimating consumption of home production in 22  GLSS3 may give rise to an offsetting bias, as the method used in GLSS3 appears to lead to a higher estimate of consumption of home production and thus agricultural income in GLSS3. The net effect of these three potential sources of bias cannot be determined. Real cocoa revenues were also declining during this period (see Table 14]. Real cocoa producer prices peaked in the 1987/88 season, which was a particularly poor harvest. Real cocoa revenues peaked in 1988/89 season, a year of record production, and then declined steadily thereafter. The only exception would have been in the northern part of Western region, which experience a 144 percent increase in production between 1987/88 and 1991/92 (Jacquet, 1995). It is difficult to link the comparison of cocoa revenues between GLSS1 and 2 to the real income trends, because it is not clear exactly which cocoa harvest households are reporting on. 5 However, it is evident from the COCOBOD data that there was a large decline in both the real producer price and real cocoa revenues between GLSS1/2 and GLSS3. The GLSS data show a similar trend: mean cocoa revenue increased slightly between GLSSI and 2, and then declined in GLSS3. Table 14. Ghana: Real Cocoa Revenue Production Real producer Real producer Real Revenue Real Cocoa Revenue Index price price index (COCOBOD) (GLSS) 1985/86 219044 6211 104 1360482 94 1986/87 227764 7506 126 1709597 118 1987/88 188171 9474 159 1782732 123 129 1988/89 300101 7933 133 2380701 165 147 1989/90 295052 6914 116 2039990 141 1990/91 293352 6266 105 1838144 127 1991/92 242807 5952 100 1445187 100 100 1992/93 212122 5556 93 1178550 82 In conclusion, what seems to be driving the increase in agricultural revenues is the sharp increase in yam production and, to a lesser extent, cassava production. Examining whether yam production actually did increase by 119 percent between 1989 and 1991 and remained high thereafter is a question worth pursuing, but beyond the scope of this paper given the data available. 5 The 1987/88 season covers the principal harvest (October 1987-Februaay 1988) and the secondary season (May -August 1988). This is essentially the same period covered by GLSS1. However households surveyed early in the survey year would more likely be reporting their revenues from the 1986/87 season, while those surveyed later in the year would more likely be reporting on revenues from the principal harvest in 1987/88. 23  Questions about the food component of the CPI index There is some reason to think that the food price index is underestimated in the period 1987/88. Table 15 compares the nominal producer price index with the food and non food indices used to derive the CPI. What is interesting is that between 1987 and 1989 the producer price indices are substantially higher than the food index, except for maize and cassava in 1989. Table 15. Ghana: Producer Price Itdices (1992=100) Nominal Producer Price Index Food Price Nonfood Total CPI Year Plantain Yam Cassava Maize Crop index Index Index Index 1985 23 20 17 20 19 23 21 22 1986 29 30 37 30 33 28 25 26 1987 49 50 77 55 62 37 34 36 1988 61 60 59 55 58 49 45 47 1989 79 85 61 48 64 61 55 58 1990 121 96 111 100 107 82 74 78 1991 95 99 100 80 93 91 90 90 1992 100 100 100 100 100 100 100 100 Note: The nominal crop index of producer price is the weighted average of norninal producer price indices of plantain, yam, cassava, and maize. Weights are the average share of consumption each crop in total rural consumption of the four crops for GLSS 1,2 and 3. Producer Price and Food Price Indices 120 100- 80 - 60 403 201 0 1985 1986 1987 1988 1989 1990 1991 1992 -4- Producer price index Food Price Index Chart 8 24  Constructing a producer price index weighted by the share of each crop's value in total consumption of the four crops for each year, we find that the producer price index is substantially higher than the food price index in 1987 and 1988 (see Chart 8). Since these commodities represent a large share of consumption, either the price index for other food commodities was substantially lower than the overall food index in the late 1980s, and thus rose rapidly between the late 1980s and 1992, or alternatively, and more plausibly, that food staples may be under-represented in the construction of the food price index used to calculate the CPI. If so, then the degree of underestimation of the CPI food index would be greatest in the 1987-88 period when real prices of maize and cassava peaked, biasing expenditure upwards in the 1987-88 period. If expenditures in GLSSI were biased upwards, the fall in expenditure between GLSS1 and 2 would tend be exaggerated. GLSS2 might then be the more appropriate baseline, with the caveat that it might represent a poor agricultural year in some areas. To get some idea of how much bias an inappropriately specified deflator might introduce, we recalculated the total deflator assuming that the food deflator were equal to the producer price index. The CPI index for GLSS1 would be 26 percent higher and expenditures for GLSSI would be 20 percent lower, bringing them below mean expenditures in GLSS2. While such a comparison is probably an extreme upper bound, it does illustrate the potential biases that could result from a misspecification of the food deflator. Trends in consumption of staple foods and producer incomes It is also usefuil to examine how the trends in consumption of the major food grain and root crops compare with the production and income trends. Table 16a indicates that consumption in rural areas, which is mostly consumption of home produced goods, fell between GLSSI and 2 and then recovered in GLSS3, though not to the same level as in GLSS L. Total consumption of maize, millet/sorghum, cassava and cocoyams was lower in GLSS3 than in GLSSI, and consumption of these cereal crops fell in GLSS2 relative to GLSS1. In contrast, consumption of yam and plantains was higher in year 3 compared with GLSS1. Table 16a. Rural Ghana: Per Capita Staple Food Consumption Millet/ Rice Maize Sorghum Cassava Yams Cocoyams Plantains Consumption of home-produced food GLSS1 732 10879 9203 14761 5248 4764 6452 GLSS2 722 7507 6900 9298 4198 3401 3923 GLSS3 1102 4538 5069 11518 6731 4424 6632 Consumption of purchased food GLSS1 3194 1949 652 4616 1514 420 1442 GLSS2 2614 1735 446 3845 1454 471 1649 GLSS3 3268 3626 1994 4749 1704 391 1305 Total consumption GLSSI 3926 12828 9855 19377 6762 5184 7894 GLSS2 3336 9242 7346 13143 5652 3872 5572 GLSS3 4370 8164 7063 16267 8435 4815 7937 25  However, these figures do not give a good idea of real consumption levels, because of shifts in relative prices. The GLSS consumption data are deflated by the overall deflator for each region, which is a weighted average of the food and non-food deflator based on food shares for each region calculated from in GLSS3. To the extent that staple food prices rose less rapidly than other food and non food prices, the procedure of deflating each commodity by the overall price deflator would tend to overestimate in the early years consumption of those commodities whose relative prices have grown less rapidly than the overall price index, and underestimate consumption of those commodities whose relative prices grew more rapidly. To get an idea of the change in real quantity consumed, we could take the nominal consumption of the commodities and deflate by the commodity specific price index. Since we only have annual producer price data not disaggregated by rural area (and since consumption, particularly of home production, is not tied to a specific reference period in GLSS1 and 2), we multiplied the commodity specific producer price index by the overall CPI price deflator to approximate nominal consumption, and then divided by the conmmodity specific price index. Table 16b shows how this would change consumption of the four commodities for which we have price data. For maize, the data still show a decline in consumption between GLSS1 and 3. For cassava consumption increases by 6 percent, plantain by 31 percent, and yam by 60 percent between GLSS1 and 3. Table 16b. Rural Ghana: Per Capita Total Staple Food Consumption Redeflated by Commodity Price Deflators Maize Cassava Yams Plantain GLSS1 1(909 15368 5260 6055 GLSS2 11215 12473 3864 4076 GLSS3 8164 16267 8435 7937 % Change GLSS1-3 -25% 6% 60% 31% % Change GLSS2-3 -27% 30% 118% 95% If we correct for change in the bias introduced by shifting from the effective to the reported months in the calculation of consumption of home production in GLSS3, consumption of home production in GLSS3 falls. This reduces the increase between GLSSI/2 and GLSS3 (see Table 16c). However, there remain fairly large increases in yam and plantain consumption between GLSS2 and 3. The increase in yam consumption is consistent with a strong increase in yam production, and also with an increase in expenditure as the elasticity of yam consumption with respect to income is higher than maize, cassava or plantain consumption. The rather large decline in maize consumption between GLSS1 and 3 is somewhat surprising. However, maize does have the second lowest expenditure elasticity of the main staple food crops (although positive), and with the fall in yam and plantain prices after 1989, perhaps consumers shifted out of maize into 26  other crops.6 If the income elasticity of demand for maize were actually negative, it might also explain why maize is the only major staple food crop in which consumption increased between GLSS I and 2 as expenditure fell.7 Table 16c. Rural Ghana: Per capita Total Staple Food Consumption Redeflated and GLSS3 Corrected for Bias in Home Consumption Maize Cassava Yams Plantain GLSS1 10909 15368 5260 6055 GLSS2 11215 12473 3864 4076 GLSS3 7619 14885 7223 7009 % Change GLSS1-3 -30% -3% 37% 16% % Change GLSS2-3 -32% 19% 87% 72% Home consumption correction factor: GLSS3 0.88 0.88 0.82 0.86 Redeflation correction factor: GLSS1 0.85 0.79 0.78 0.77 GLSS2 1.21 0.95 0.68 0.73 6 See Alderman and Higgins (1992) for a discussion of income and price elasticities. 7 There was also a change in the questionnaire regarding the various components of maize consurnption, but this change does not appear to be significant. GLSS1/2 asked two questions, one about maize (cob, grain, or flour) and one about kenkey. GLSS3 asked two questions one about maize (cob-fresh) and the other maize (flour/dough), but there was no question about consumption of home-produced kenkey. However, kenkey was not a large expenditure item in the CHP section in GLSS1/2. 27  4. ACCOUNTING FOR THE VERY LARGE REDUCTION IN POVERTY IN THE SAVANNAH The headcount index for the Savannah region fell by some 11 percentage points between 1988 and 1989, and 16 percentage points between 1989 and 1992, mostly due to the growth in mean expenditures. While both food and nonfood expenditures in GLSS3 appear to be biased upwards for all rural areas, there are some reasons to think that the growth in expenditure and consequent reduction in poverty is exaggerated for the Savannah than for other two localities for two reasons. First, the GLSS3 expenditure data seem more biased for the Savannah than for the other two localities. Second, the 1989 - 1992 comparison may provide an overly optimistic view of the growth in expenditures, because 1989 was a worse agricultural year in the Savannah than in the other two rural localities. The Savannah region had by far the largest growth in total food expenditures between 1988 and 1992, almost double that of the rural coastal region. While there was a large drop in the food share in other regions between 1988 and 1992, there was virtually none at all in the Savannah, despite the fact that total expenditures increased and the relative price of food fell. This suggests that it may be worth looking into the large increase in Savannah food expenditures (see Table 4). The Savannah had a particularly largely increase in several categories of real food expenditure between GLSS2 and 3 (see Table 17). Pulses increased by roughly 7,300 cedis, largely due to spectacular increase in consumption of home-produced groundnuts. Other categories that showed a particularly dramatic increase in oilseeds (roughly 2,800 cedis), poultry (4,300 cedis), vegetables (2,200), and alcohol (3,800 cedis). Whether these increases are "real," or whether they are due to large relative price shifts or to changes in the questionnaire or sampling bias is not clear. However, the increase in groundnuts is particularly suspect. According to the FAO data, the 1991 groundnut harvest was about a third of the 1987 harvest, and the Savannah is essentially the ordy region in which groundnuts are grown. Groundnut prices would have had to have skyrocketed to account for the huge increase in the value of groundnut consumption. If the price increase were that large, it should have had an important effect on the food deflator for the Savannah region. We do not know whether the rural CPI adequately captures changes in the prices across regions. 28  Table 17. Rural Ghana: Per Capita Food Consumption by Categories* Pulse Oil- Oils/ Fruits Vege- Meat Poultry Milk Beve- Misc Alcohol Cereal Starch Fish Sugar Spice Prepared TOTAL seeds Fats tables rage food ..CoastaI.................................................................................................................................................................................................................................................................................... Rural Coastal GLSS1 2889 4261 3067 2855 12307 2852 2793 1367 658 683 3188 19009 34780 23155 2762 943 11672 129239 GLSS2 2355 3252 2871 1862 11639 2514 2730 1509 680 933 2848 15662 25431 26665 2272 945 8467 112634 GLSS3 2190 3946 4755 2046 11462 3203 2991 1177 1689 288 5892 18501 33982 28665 1949 2259 9837 134832 % Change GLSSI-3 -24% -7% 55% -28% -7% 12% 7% -14% 157% -58% 85% -3% -2% 24% -29% 140% -16% 4% Rural Forest GLSSI 2865 2541 2825 3817 13752 5918 2816 1086 196 556 3005 15609 48157 19345 1844 1246 6599 132177 GLSS2 2816 1794 2920 2893 12697 4829 2736 870 328 668 2670 11776 34899 22613 2099 1176 8049 115830 GLSS3 2049 2900 4363 2296 11293 6717 4121 830 1323 224 4087 11158 45991 20928 1538 1962 6539 128319 % Change GLSS1-3 -28% 14% 54% -40% -18% 13% 46% -24% 575% -60% 36% -29% -4% 8% -17% 57% -1% -3% Rural Savannah GLSSI 5735 1158 1893 1162 10442 3271 1377 622 251 83 2653 53647 30676 8764 1855 1360 2498 127448 GLSS2 4198 807 2058 1123 8882 3361 1319 463 132 57 3542 42799 21649 8831 1777 1414 3723 106136 GlSS3 13055 3995 3187 990 12595 4317 5696 707 517 260 6482 39700 29160 9216 1495 3054 3630 138056 % Change GLSSI-3 128% 245% 68% -15% 21% 32% 314% 14% 106% 213% 144% -26% -5% 5% -19% 125% 45% 8% * Excludes in-kind payment for employment. 29  Moreover, the discrepancy between the reported months of consumption of home production and the effective months is particularly large for the Savannah primarily because of the large discrepancy for millet/sorghum (see Table 18). If we correct the consumption of home production of staple food crops by the ratio of mean reported months to effective months, consumption of home production in GLSS3 would drop by almost 14,000 cedis. This is equivalent to the increase in total food expenditure between GLSS1 and GLSS3. While a fall of this magnitude is probably not plausible, it does raise some doubts about the comparability of the data. Table 18. Rural Savannah: Per Capita Consumption of Home-produced Staples Consumption of home-produced staples .....................................I....................................................................- .................. .................. ............................................... .................................................................. .............. Rice Maize Millet! Cassava Yams Cocoyams Plantain TOTAL Total Sorghum Expenditure GLSS1I 1592 17769 27004 12381 10516 1770 1676 72708 167086 GLSS2 1526 13068 21143 6391 8279 1902 957 53266 144918 GLSS3 2744 6382 14684 5552 13375 1296 2358 46391 188469 GLSS3 Revised* 1683 4595 8957 4775 9229 1115 2122 32476 * GLSS3 correction factor for consumption of home-produced staples Rice Maize Milletl Cassava Yams Cocoyams Plantain Sorghum 0.61 0.72 0.6:1 0.86 0.69 0.86 0.90 On the agricultural side, there is reason to think that the very large decrease in poverty in the Savannah between 1989 and 1992 was due to a rebound in combined millet and sorghum production between 1988 and 1991 harvest, in addition to the increase in yam production. The Savannah consumes twice as much yam per capita as the rural coastal and rural forest regions combined, and thus it stands to reason that if yam production did show a spectacular increase from 1991 onwards, that there would have been a large reduction in poverty in the Savannah. Also the largest increase in percentage of the population engaged in retail trade (virtually all women) took place in the Savannah, suggesting that the increase in yam production may have fostered an increase in trade (see Table 19). 30  Table 19. Rural Ghana: Retail Trade Nonfarm Self Employment ............................................................................................................................. .......... ....................................................................................... Percent of Women Percent of Self-employed Mean Percentage in Retail Trade Population in Retail Trade Self-employment Income from Retail Trade ................... ....................................................................................................................................................................................................... Rural Ghana GLSSI 82.5 39.8 33.1 GLSS2 85.8 40.0 32.9 GLSS3 88.9 50.1 42.5 Rural Savannah GLSS1 83.0 30.9 25,3 GLSS2 81.7 28.4 23.0 GLSS3 87.2 55.0 48.1 In addition, the reduction in poverty between GLSS2 and GLSS3 may also be magnified by the apparent poor harvest of sorghum and millet (crops basically grown only in the Savannah region) in 1988. The fall 1988 harvest would have affected crop revenue and consumption in 1988/89 GLSS2 survey. According to the FAO data, the millet harvest was down by 34 thousand tons (a drop of 20 percent), while the sorghum harvest was down by 45 thousand tons (a decline of 22 percent). The GSS data are more optimistic: they show the 1988 sorghum harvest falling by 28 thousand tons, with an increase in millet production of 19 thousand tons. If the production trends are in line with the FAO data, they would explain the severe increase in poverty in the Savannah region in GLSS2. And by the same token, it would also be misleading to take the GLSS2 data as a baseline, since doing so would give an overoptirnistic view of the reduction in poverty between 1989 and 1992. So for the Savannah, at least, GLSSI might be the better baseline, depending on the magnitude of the bias, if any, introduced by an improperly specified food deflator. 31  E. CONCLUSION Thus by and large the evidence seems to be suggesting that there was a large incr.ease in production of non cereal staple food crops in the early 1990s. Agricultural proclucer incomes likely increased between 1989 and 1992. Some increase in total expenditures would thus be consistent with this trend. However, there is also reason to believe that the GLSS data strongly overstate the increase because of upward biases in the GLSS3 data relative to the GLSS1/2 data due to changes in the questionnaire. The GLSS expenditure data for the Savannah seem particularly suspect, given some very large increases in certain consumption items. To the extent that there was a significant increase in yam (and perhaps cassava) production, it would not be surprising to see a strong growth in rural non farm income. The data indicate that much of the growth came in retail trade by women. Other accounts also suggest that there has been a major increase in yam marketing over the decade. If so, this explanation opens up a whole new puzzle: accounting for the spectacular increase in root crop production in the absence of any apparent technological improvement or major infrastnucture improvements. One possible explanation might be that households shifted out of cocoa and into yam production, particularly in areas where there are problems with cocoa diseases. Since the GLSS3 data suggest a strong increase in yam consumption in the Savannah (and a strong increase in yam production), which is not a cocoa growing region, this explanation would not seem to account fully for the increase in yam production Further research is needed to ascertain whether there was a major increase in the production of yam and other non cereal staple food crops, and if so, why. To the extent that there was a reduction in poverty based on the increase in agricultural production, it is worth asking the question of the extent to which the major policy changes undertaken in Ghana were responsible for the increase. As it appears that nontradables crops benefited the most, the real devaluation of the cedi in the mid 1980s would not seem to be a major factor in agricultural growth and poverty reduction. Thus, while casting serious doubt on the comparability of the GLSS data over time, this paper suggests that we need to look at what happened to the agricultural sector in Ghana to understand what role-if any-policy changes may played in what is most likely a smaller reduction in poverty than what is suggested by the GLSS data. 32  References Alderman, Harold, and Paul Higgins. 1992. "Food and Nutritional Adequacy in Ghana." Cornell Food and Nutrition Policy Program Working Paper 27. Ithaca, N.Y.: Cornell University. Coulombe, Harold and Andrew McKay. 1995. "An Assessment of Trends in Poverty in Ghana, 1988-92." PSP Discussion Paper, The World Bank, Washington, D.C. (November). Datt, Gaurav and Martin Ravallion. 1992. "Growth and Redistribution Components of Changes in Poverty Measures. A Decomposition with Applications to Brazil and India in the 1980s." Journal of Development Economics, 38 (1992) 275-295. Jones, Christine and Xiao Ye. 1995. "Understanding Poverty Trends in Ghana, 1988- 1992." Draft for the Ghana Extended Poverty Study, The World Bank. Leenhardt, Blaise. 1993. "Prix des Vivriers et Biais Urbain au Ghana." Caisse Francaise de Developpement. Processed. Jacquet, Laurent. 1995. "Le cacao demeure la deuxieme ressource du Ghana." Marches Tropicaux, 722, April 7, 1995, pp722-725. Scott, Chris, and Ben Amenuvegbe. 1990. Effect of Recall Duration on Reporting of Household Expenditures: An Experimental Study in Ghana. Social Dimensions of Adjustment Working Paper No. 6, World Bank, Washington D.C. World Bank. 1995a. A Synthesis of the Extended Poverty Study. Report No. 14504- GH. Washington, D.C.: The World Bank. World Bank. 1995b. Ghana: Growth, Private Sector, andPoverty Reduction -- A Country EconomicMemorandum. Report No. 14111- GH. Washington, D.C.: The World Bank. 33   Annex Table Al. GHANA: Crop Production Statistics, Reported by Ghana Statistical Service Year Yams Cassava Plantain Cocoyam Rice Millet Sorghum Maize 1980 525000 2896000 931000 848000 64000 136000 156000 354000 1981 463000 2721000 835000 972000 44000 131000 171000 334000 1982 374000 1986000 763000 756000 37000 120000 126000 264000 1983 354000 1375000 755000 613000 27000 114000 106000 141000 1984 725000 4005000 1234000 2835000 76000 139000 176000 574000 1985 660000 3075000 1360000 900000 80000 120000 185000 395000 1986 1048000 2876000 1088000 1005000 70000 110000 128000 559000 1987 1185000 2725000 1079000 1012000 81000 173000 206000 597000 1988 1200000 3300000 1200000 1115000 105000 192000 178000 600000 1989 1200000 3320000 1040000 1200000 67000 180000 215000 715000 1990 877000 2717000 799000 815000 81000 75000 136000 553000 1991 2632000 5701000 1178000 1297000 151000 112000 241000 932000 1992 2331000 5662000 1082000 1202000 132000 133000 259000 731000 1993 2720000 5973000 1322000 1236000 157000 198000 328000 961000 1994 1700000 6025000 1475000 1148000 162000 168000 324000 940000 Table A2. GEIANA: Crop Production Statistics, Reported by FAO Year Yams Cassava Plantain Cocoyam Rice Millet Sorghum Maize 1980 650000 1857600 734000 643000 78000 82000 132000 382000 1981 591000 2065000 829000 631000 97000 119000 131000 378000 1982 588000 2470000 745000 628000 36000 76000 86000 346000 1983 866000 1720000 755000 720000 40000 40000 56000 172000 1984 1178000 2200000 1400000 800000 65000 133000 172000 696000 1985 987000 2300000 1629000 900000 80000 112000 145000 584000 1986 1048000 2876000 1087000 1005000 69600 110000 128000 559100 1987 1185400 2725800 1077600 1011800 80700 173100 205900 597700 1988 1200000 3300000 1200000 1115000 95000 139000 161000 751000 1989 1280000 3327200 1040000 1200000 73700 180000 215000 715000 1990 877000 2717000 799000 815000 80900 74500 135800 552600 1991 1000000 3600000 1178300 1296800 150900 112400 241400 931500 1992 1000000 4000000 1082000 1202200 131500 133300 258800 730600 1993 1000000 4200000 1321500 1235500 157400 198100 328300 960900 34