Adjustment with a Human Face? Evidence from Jamaica

World Development Vol. 31, No. 7, pp. 1125–1145, 2003 Ó 2003 Elsevier Science Ltd. All rights reserved Printed in Great Britain 0305-750X/03/$ - see front matter

www.elsevier.com/locate/worlddev

doi:10.1016/S0305-750X(03)00063-9

Adjustment with a Human Face? Evidence from Jamaica SUDHANSHU HANDA Inter-American Development Bank, Washington, DC, USA and DAMIEN KING * University of the West Indies, Mona, Jamaica Summary. — In September 1991 Jamaica liberalized its exchange rate as part of its Structural Adjustment Program (SAP). The sudden and steep devaluation associated with this policy move had serious repercussions for real purchasing power, poverty, general inflation, and food prices, especially staple foods such as rice and flour which are imported. This study evaluates the ‘‘social cost’’ of the liberalization policy by examining the behavior of preschool childrenÕs weight for height or wasting, an indicator of nutritional status that is sensitive to short term fluctuations in living conditions. Using 8 years of national microsurvey data for 1989–96, we apply ‘‘synthetic cohort’’ analysis to disentangle the separate impacts of childÕs age, date of birth, and measurement date, on weight for height. Estimates based on an exhaustive set of controls indicate that children weighed in the aftermath of the policy (November and December 1991) are significantly lighter (by 0.178 z-scores) than children weighed just a few months later, and children in urban areas were especially affected. When food price inflation is explicitly entered into the model, it is highly statistically significant, and reduces (but only slightly) the effect of being measured at the end of 1991. The calculated elasticity of weight for height z-score with respect to food price inflation is a very high )0.86. During the rapid economic reform in 1991, this elasticity rose to )1.24, indicating a large response of weight for height to food price inflation, as is to be expected in a small open economy. Ó 2003 Elsevier Science Ltd. All rights reserved. Key words — nutrition, structural adjustment, devaluation, Jamaica

1. INTRODUCTION The social impact of structural adjustment policies (SAP) in developing countries is a controversial issue. These policies, typically consisting of reforms such as trade and exchange rate liberalization, government sector retrenchment, and removal of food and other consumer subsidies, have been associated with sudden increases in poverty and income distribution in Latin American and the Caribbean (Berry, 1995; Handa & King, 1997). On the other hand, Behrman and Deolalikar (1991) argues that trends in human resource indicators such as health, nutrition and education do not show a clear deterioration in the face of SAP, despite claims to the contrary in Jolly, Stewart, and Cornia (1984).

Studies attempting to estimate the impact of SAP on social outcomes such as poverty and inequality face two challenging tasks. First, establishing causality involves isolating the impact of specific adjustment policies on

* This

research was funded by the FOCAL project of the Center for International Studies, University of Toronto, Toronto, Ontario, Canada. Thanks to Nicola Martin, Brian Langrin, and Helder Zavale for excellent research assistance; Albert Berry, Gerry Helleiner, Kristin Fox, Patricia Mucavele and John Hoddinott for stimulating discussion; and to seminar participants at the IDB/IFPRI conference on Shocks for useful comments. The opinions in this paper are not necessarily those of the IDB. All errors are the authorsÕ responsibility.

1125

1126

WORLD DEVELOPMENT

human development outcomes and is extremely difficult, not simply because of the multitude of potential intervening factors which are hard to control in standard regression analysis, but also because the effects of these policies generally occur with a lag, and the lag time itself will depend on the sequence of policy changes as well as institutional and structural features of the economy. 1 Second, even if a causal link can be established between particular market-friendly reforms based on SAP and a deterioration in social indicators, the (almost unanswerable) question remains as to whether this deterioration would have been less without the reform package, since in many cases SAP are part of International Monetary Fund (IMF) sponsored loan agreements negotiated by economically distressed countries with severe balance of payments crisis. The present study contributes to the on-going debate on the social consequences of SAP by analyzing the behavior of childrenÕs weight, a nutritional indicator that responds to short run living conditions, around the period of JamaicaÕs exchange rate liberalization in 1991. For the purposes of our statistical analysis, the liberalization episode in Jamaica is very close to a ‘‘natural experiment’’ in that it occurred at a very specific time, with very specific consequences that can be formally linked to food prices and hence the availability of nutrients and other anthropometric inputs. Furthermore, we use a unique set of eight annual household surveys where we are able to exploit ‘‘synthetic cohort’’ estimation techniques popular in the immigration literature (see, for example, Baker & Benjamin, 1994). This technique allows us to decompose variations in childrenÕs weight into age effects, cohort or birth year effects, and year or time effects which capture common macroeconomic conditions. 2 By identifying children who are weighed within a certain window of the macro ‘‘shock,’’ we are able get a better measure than most previous studies on the possible impact of the liberalization reform on one particular social indicator. As a starting point, this study extends the work of Behrman and Deolalikar (1991) by ‘‘establishing the facts’’ of whether short-run child anthropometry suffered in Jamaica during a period of high inflation induced by exchange rate liberalization, although the extent of SAP economic reform is much greater during our period of study than that of Behrman and Deolalikar. Furthermore, because of the richness of our data and the advances in method-

ology as compared to previous work in this area, we are in a better position than most to establish causality between SAP and human development. In particular, because we are able to compare children weighed during the window of the macro shock with other children, and given that inflation-inducing devaluation was the main economic reform during this time period with direct consequences for the poor, our estimates can be interpreted as capturing the impact of this reform on short-run nutritional status. The results of the analysis show that despite JamaicaÕs very high level of human development relative to other countries with the same level of per capita income, the z-score of weight for height of preschool children fell by 0.178 zscores in the months immediately after the devaluation; the drop in weight was especially great among urban children. The results also indicate a very strong relationship between child weight, exchange rate devaluation, and food price inflation. The estimated point elasticity of the z-score with respect to the exchange rate and food price inflation is )0.36 and )0.86, respectively, and the short-run food price elasticity during the liberalization rose to )1.24. This strong link between currency depreciation, inflation, and child welfare, illustrate the political and economic sensitivity of the exchange rate in open economies such as Jamaica, with clear implications for the timing and manner in which SAP reforms should be implemented. 2. STRUCTURAL ADJUSTMENT, MACROECONOMIC REFORM, AND LIVING STANDARDS Jamaica has had a long history of SAP-type reforms, dating at least from the 1977 IMF Stand-by agreement. As demonstrated in Handa and King (1997) and others, however, Jamaica was essentially a late reformer, as the implementation of adjustment and stabilization had been sporadic and inconsistent until the end of the 1980s, when there was a sudden acceleration in the pace and consistency of economic reform. 3 While the beginning of the 1990s saw profound reforms in all of the typical components of the structural adjustment package (government reduction, domestic market liberalization, external openness, and stabilization), we are particularly interested in monetary policy and its impact on the exchange rate and domestic price inflation.

ADJUSTMENT WITH A HUMAN FACE?

At the beginning of the period of serious economic reform, 1989, some degree of monetary stability had been achieved with three consecutive years of single-digit inflation during 1986–89. In 1988 and 1989, the central bank accommodated fiscal spending occasioned by a natural disaster (Hurricane Gilbert). The effects of this on credit expansion were multiplied in 1990 and 1991 by financial liberalization, which freed the financial sector to reallocate its credit portfolio. Financial liberalization began with the removal of exchange controls early in the decade. In 1990, commercial banks were authorized to transact in foreign exchange and foreign exchange accounts were permitted. In 1991, exchange controls were removed. Until 1993, however, the central bank used indicative exchange rates and moral suasion in commercial banksÕ foreign exchange operations. Initially, surrender requirements were imposed on all institutional foreign exchange transactions. These were gradually reduced until they were discontinued in 1996. As a result of both of the above factors, M1 grew by 28% in 1990 and by 95% in 1991; by 1990 the annual inflation rate reached 30%, and in 1991 it jumped to 80%, an all time high. The monetary expansion of 1988–91, along with the domestic inflation that it generated, served to bring about a sharp depreciation of the currency immediately upon the removal of exchange controls in 1991; the value of the Jamaican relative to US dollar went from seven in 1990 to approximately 12 in 1991 and 22 in 1992. Figure 1 shows the movement (measured by the percentage change over the previous six months) in the exchange rate and CPI during the period of study; the spike in the exchange

1127

rate corresponds to the period that includes September 1991, when the exchange rate was fully liberalized. Inflation also peaked at this time, but both the depreciation of the currency and the increase in the CPI began in the latter part of 1990. Since the trade adjustment effects of a depreciation tend to occur gradually (and in any case, one of JamaicaÕs two main exports, bauxite, is not labor intensive), the main effect of the depreciation on the poor and middle class in the short run is through higher import prices. This is especially significant in Jamaica since staple foods (such as rice and flour) are imported, hence the devaluation worsened the position of poor and vulnerable households in terms of access to basic foods and hence nutrients. Table 1 shows the evolution of poverty and food prices around the period of exchange control liberalization and there is every indication of a sudden loss of purchasing power of Jamaican households––the poverty rate jumped by 15 percentage points during 1991–92. The food component weight in CPI is 55%, and during this period the average contribution of food price inflation to overall inflation was 61% (Economic & Social Survey of Jamaica, 1996). Table 1 and Figure 2 show the point-to-point increase (%) in prices of the total food component of the CPI, as well as the three major components of the food basket that account for important staple food consumption in Jamaica: starches (rice, yam and cassava), baked products (bread), and meats (chicken and cod fish). In all cases there is a large spike in these series for 1991, although inflation was significantly lower in the year after liberalization (1992).

120 Exchange Rate CPI

100 80 60 40 20 0 89(6)

90(6)

91(6)

92(6)

93(6)

94(6)

95(6)

-20

Figure 1. Exchange rate and CPI in Jamaica.

96(6)

97(6)

1128

WORLD DEVELOPMENT Table 1. Real wages, poverty, and food prices in Jamaica 1988–96

Year

1988 1989 1990 1991 1992 1993 1994 1995 1996

Poverty rateb

Real wage indexa

100.2 100.0 97.2 82.8 69.1 89.4 99.5 100.9 110.1

45.5 47.3 44.6 60.0 43.2 45.9 50.1 54.5

Point-to-point change (%) in price ofc All food

Starches

Baked products and cereal

Meat, fish and poultry

10.6 20.9 29.0 84.3 40.4 31.8 27.1 27.6 12.1

36.5 16.4 8.30 49.8 39.4 36.5 40.0 32.5 9.3

Not avail. 26.6 29.1 96.7 48.1 18.2 32.8 33.3 13.4

2.3 22.3 38.4 92.2 38.4 29.4 16.0 26.6 9.0

a

STATIN (various years) earnings and employment in large establishments. Calculated by authors from Survey of Living Conditions, maintaining US$60 per month in 1989 as real poverty line and consumption per capita as the welfare measure. c Economic and Social Survey of Jamaica, (various years). b

100 90

All Food Starches Baked Products & Cereal Meat, Fish & Poultry

80 70 60 50 40 30 20 10 0 1988

1989

1990

1991

1992

1993

1994

1995

1996

Figure 2. Point-to-point inflation of food and components.

While the impact of financial liberalization on inflation and wages was dramatic in 1991, the effect appears to have been concentrated and short. For example, both the labor force participation rate (44%) and the unemployment rate (16%) remained constant during the period, while the number of hours worked increased significantly in 1991 and then returned to its pre-shock level in 1992 (STATIN, various years). Meanwhile, real per capita GDP growth continued to be positive, dropping from 5% to 0.5% during 1990–91 and increasing to 2% in 1992 and 1993. The real wage reported in Table 1, which reflects primarily formal sector em-

ployment where wages are governed by fixed contacts, declined significantly in 1991 and 1992, but rebounded in 1993 and regained its pre 1991 level by 1994. But the minimum wage, which governs the pay of hourly or short-term employees, actually increased significantly (by 20%) during 1991–92 (Alleyne, 2001), indicating that the economic shock may have had more of an effect on the middle class who tend to work in the formal sector governed by relatively inflexible wage contracts, than on the poor in flexible contracts whose wages are closer to the officially sanctioned minimum wage.

ADJUSTMENT WITH A HUMAN FACE?

Overall, given the structure of production and consumption of tradables, the impact of the sudden currency devaluation in 1991 dramatically decreased real incomes in Jamaica. With imported food staples suddenly more expensive, unless there was a large degree of consumption switching, the availability of nutrients is likely to have declined significantly, with implications for the short run nutritional status (wasting) of pre-school children. But the duration of the impact of the shock seems to have been short and concentrated. Employment remained steady, hours of work increased, growth in total output continued to be positive, and the minimum wage actually increased significantly, which may have served to ‘‘protect’’ the poor relative to the middle class.

3. THEORETICAL CONSIDERATIONS AND LITERATURE REVIEW That shallowness of financial markets is a characteristic of less developed economies was established at least as long ago as the work of Shaw (1973) and McKinnon (1973). The manifestations are low, often negative interest rates, a paucity of savings and insurance instruments, and scarcity of channels. Even in economic regimes that do not engage in repressive financial policies, the formal financial market institutions do not reach the poorest households. Thus, the ability of poor households to achieve consumption smoothing in the face of economic shocks is, theoretically at least, limited. But, the experience in many less-developed countries (LDCs) is that a vast array of behavioral and institutional responses arise, from informal credit markets to extended family networks, to fill the need to effect consumption smoothing. It then becomes an empirical exercise to determine how complete those responses are, and therefore how effective are the options for achieving consumption smoothing. A sizable literature exists on the extent to which households in poor countries are successful at this objective. Paxson (1992) finds that the poor do try to take advantage of whatever mechanisms of risk management are available, informal or otherwise. Townsend (1995) uses the ICRISAT data––a panel from three south Indian villages drawn over 10 years––to show that household income varies over time from the village average. Townsend (1995) concludes that there is considerable but

1129

incomplete usage of insurance. Note that most of this work uses data from rural households. The nature of the consumption smoothing mechanisms available suggest why poor households do worse. Generally, consumption smoothing may be achieved either by opting for less volatile income streams or by accessing credit and insurance to smooth consumption in the presence of a given variable income. A reduction in income volatility may be pursued by means of diversification, which in the farm context means a planting a variety of crops or having some nonfarm income through employment off the farm (Morduch, 1995). Home production is also a means of risk management in this context. Once the income stream has been determined, households may turn to various institutions of credit and insurance. Ownsavings is the most obvious means of insurance. For poor households, however, they have little savings by definition, and frequently lack access to good savings instruments that offer a dependable store of value. With the volatile and moderate to high inflation that often characterizes developing economies, money is a poor store of value. Farm households have the option of stockpiling produce, but their product is perishable. Rotating savings and credit associations are also a popular means of insurance insofar as they allow out-of-turn withdrawals. Jacoby and Skoufias (1997) find that child labor plays a significant role as a means of selfinsurance. In the absence of an adequate means of insurance, households turn to credit. In general, commercial credit is inaccessible, so families may resort to informal lenders. The extended family is also an important source of credit as well as insurance. In summary, the mechanisms for consumption smoothing reviewed above are best suited to rural households facing idiosyncratic shocks. Urban households are more likely to be disassociated from the extended family and do not have the option of crop variety, while aggregate shocks are inherently more difficult to insure against. The present study looks at the effects of an aggregate shock and includes a significant portion of urban households in the sample; this is the setting where informal mechanisms of consumption smoothing are least likely to be effective. Looking at the effect of an income shock on children represents the extremity of failed consumption smoothing, on the presumption that caregiving adults attempt to absorb as much of the shock as possible before depriving their

1130

WORLD DEVELOPMENT

children. Many have shown the effects of shocks on children, including Behrman (1988), who looked at the extent to which the health of children suffer during the harvest in rural households in south India, Foster (1995), who compared body size changes before and after flooding in Bangladesh, and Jacoby and Skoufias (1997), who demonstrated that school attendance was related to adverse income shocks in rural south India. These exercises all related, however, to the children of rural households. The inability to mitigate income shocks is important, quite apart from the human suffering that it entails. As has been argued and demonstrated by Morduch (1995), if insurance is ineffective, agents will choose low volatility employment and production techniques instead of the most profitable and wealth-creating options, lowering living standards and slowing technological adoption and economic growth. Morduch has estimated that households may be willing to give up 16% of expected income in pursuit of their preference for a smoother stream of income, all because of an absence of options for smoothing consumption after the income stream has been determined. 4. DATA AND EMPIRICAL FRAMEWORK (a) Data Data for this study come from the Jamaican Survey of Living Conditions (SLC), an annual multi-purpose national household survey collected by the Statistical Institute of Jamaica. The SLC contains modules on consumption expenditure, health, education, housing, and social services, with a typical sample size of 2,000 households and 8,000 individuals (although the 1989 and 1992 rounds surveyed twice as many households than usual). The health module contains a section for children up to 60 months of age with information on anthropometry, immunizations, birth weight, and date of birth, while the cover module contains information on region of residence and the date the child was measured. We use the 1989–96 rounds of the SLC to assess the trend in weight for height. A time-series of cross-sections as we have here allows us to trace the welfare of cohorts rather than individuals as in a regular panel or longitudinal data set, where cohort is defined

according to birth year. A time series of crosssections has several advantages over a panel. For our study, the key benefit is that because fresh samples are selected each year, the common attrition problem in regular panels is minimized, provided the sampling frame is updated regularly and there is no serious migration or mortality. 4 In the set of surveys used in this study, the age distribution of preschool children is approximately the same across survey years, indicating that attrition bias (or sample inconsistency) is not a problem in the SLC data. (b) Empirical framework To describe the evolution of weight for height in Jamaica while at the same time accounting for possible fluctuations induced by macroeconomic conditions, we relate current child nutritional status to three sets of factors Ni;t ¼ ci þ yt þ ðxi;t þ ui;t Þ

ð1Þ

Current nutritional status of child i in time period t (Ni;t ) depends on a child-specific fixed effect such as the date of birth ci , 5 current economic conditions, some of which include the effects of macroeconomic policy reform but also secular trends and seasonality (yt ), and child-specific factors that can vary with time, some of which are observable (xi;t ) such as access to and quality of health care, age, sex, household resources and characteristics, and others that are unobservable (ui;t ). Alternatively, (1) can be interpreted as the reduced form relationship (‘‘demand function’’) for child health derived from a standard new household economics model stressing household or home production of nonmarket goods such as childrenÕs health and education. Such models are described in detail by Strauss and Thomas (1995) and Behrman and Deolalikar (1988), and applications can be found in Handa (1999); Alderman and Garcia (1994); and Strauss (1990). Based on (1), our estimating equation becomes Ni;t ¼ a þ B1 c þ B2 y þ B3 xi;t þ uit

ð2Þ

where a is a constant, the BÕs are the parameters, uit is an individual specific random error for child i (in time period t), and the sample consists of individual observations on these variables and the z-score 6 of weight for height (Ni;t ) taken from eight rounds of the SLC. In addition to weight for height we also provide some exploratory estimates using height for age

ADJUSTMENT WITH A HUMAN FACE?

to see whether there are any long-run consequences of the 1991 liberalization episode. An important empirical issue is controlling for the established growth patterns of young children. Following the recommendation of nutritionists, children are divided into eight age groups and dummy variables included to allow for flexible estimation of age effects on growth. The age groups (in months) are 0–2, 3–5, 6–8, 9–11, 12–23, 24–35, 36–47, and 48–60. Aside from age, we also include sex of the child, 13 parish dummy variables and two dummies for residence in a rural or urban area, these latter variables to capture living conditions and access to health and other social services. To control for differences in caregiving behavior and knowledge of health issues, years of schooling of the primary caregiver and her age are included in the estimation. For the 1989, 1990, and 1991 surveys the caregiver is the mother. For the other years it was not possible to identify the childÕs mother so the education of the oldest female resident is used instead. Finally, as a measure of household resources we include the log of per capita consumption expenditures, instrumented using type of dwelling (house, apartment, semi-detached, etc.), main material of walls, water supply, kitchen and toilet facilities, ownership of house, access to electricity, and telephone. The cohort effects (c) are captured by the childÕs date of birth using six-month (bi-annual) dummy variable, 7 and the common macroeconomic effect (y) is the time of the survey. We use two definitions of y, starting with year of survey, and then the date of the actual measurement of the child since in each survey year the anthropometric data was collected over a

1131

period of several months (see below). We then explicitly measure the impact of macroeconomic conditions on child weight by including measures of inflation and currency depreciation in the model. (c) Descriptive statistics Tables 2 and 3 provide the mean z-score of children in our sample by year (Table 2) and age group (Table 3). We are particularly interested to see if there is a ‘‘blip’’ in the data in 1991 when Jamaica suffered severe macroeconomic adjustment with deleterious consequences for income and food prices. Table 2 shows that for all children, mean z-score of weight for height declined from )0.030 in 1989 to )0.17 in 1990, and then to )0.30 in 1991, returning to normal in 1992. To allow for differences in vulnerability due to age, Table 3 provides mean z-scores by age group. Comparing each age group across the surveys (i.e. going across the rows), we see that in four of the six age groups the mean weight for height zscore is lowest for survey year 1991; in the two cases where it is not (age groups 6–11 and 48– 60), the lowest mean weight for height z-score occurs in 1990, when the inflation rate began to increase (see Figure 1). Our expectation regarding the pattern of stunting (height for age), shown in the bottom half of Table 3, is less clear. If stunting developed due to growth faltering caused by extreme adjustment in 1991, we would expect this to show up in the data some 6–18 months afterward. Considering children age six months and over after 1991 (those 0–5 months after 1991 are not in the 1991 survey) we find that four of the five age

Table 2. Mean of anthropometric indicators for all pre-school children Weight for heighta

Year

a b

Height for ageb

Mean

Observations

Mean

Observations

1989 1990 1991 1992 1993 1994 1995 1996

)0.030 )0.170 )0.302 )0.129 )0.053 )0.035 )0.092 )0.128

1,146 668 622 1,499 726 739 769 729

)0.300 )0.169 )0.228 )0.262 )0.312 )0.273 )0.166 )0.164

1,139 653 613 1,466 726 727 763 730

All

)0.102

6,898

)0.153

6,817

Children with z-scores greater than 5 or less than )5 excluded. Children with z-scores less than )5 or greater than 3 excluded.

1132

WORLD DEVELOPMENT Table 3. Mean anthropometric status by age group and survey year

Year: age group

1989

1990

1991

1992

1993

1994

1995

1996

Obs.

Weight for height z-score 0–5 0.419 6–11 0.182 12–23 )0.140 24–35 0.037 36–47 )0.051 48–60 )0.289

0.539 )0.148 )0.088 )0.294 )0.287 )0.361

0.054 0.071 )0.372 )0.354 )0.412 )0.328

0.404 0.087 )0.118 )0.194 )0.194 )0.240

0.549 0.375 0.072 )0.343 )0.036 )0.394

0.616 0.303 0.158 )0.155 )0.151 )0.070

0.577 0.084 )0.044 )0.214 )0.261 )0.259

0.649 0.301 )0.234 )0.271 )0.195 )0.282

531 645 1,477 1,439 1,455 1,351

Height for age z-score 0–5 0.255 6–11 )0.165 12–23 )0.478 24–35 )0.627 36–47 )0.116 48–60 )0.302

0.443 )0.238 )0.526 )0.096 )0.155 )0.073

0.346 )0.343 )0.527 )0.023 )0.159 )0.357

0.170 )0.463 )0.482 )0.156 )0.095 )0.373

0.516 )0.376 )0.818 )0.337 )0.074 )0.284

0.080 )0.121 )0.613 )0.285 )0.269 )0.083

0.383 )0.255 )0.498 )0.164 )0.051 )0.129

0.069 )0.233 )0.554 0.076 )0.027 )0.155

518 635 1,447 1,419 1,446 1,352

groups report the lowest z-score in either 1992 or 1993, the two years immediately proceeding the shock. Figure 3 graphically displays some of the numbers from Table 3, comparing mean nutritional status by age group for three survey years, 1989 (before), 1990–91 (during), and 1993–94 (after); the graph shows that the 1990– 91 sample of children have the lowest mean z-scores at all age groups. 8 Behrman and Deolalikar (1991) 9 analyze the trend in child nutritional status in Jamaica during 1977–87 using annual averages calculated from clinic-based data. They find no significant downward trend in average nutritional status over that period, and conclude that the structural adjustment that occurred in Jamaica

during that period did not have a significant negative social impact. Using our individual level data for 1989–96, we estimated equations analogous to the ones found in Behrman and Deolalikar (1991), where weight for height was regressed on xit (that is, sex, age group dummies and the 13 parish and two regional indicators), and a time trend to characterize the secular path of the dependent variable. We looked for two possible deviations in the data. The first was a one-shot deviation in 1991, captured by a dummy variable indicating whether the observation came from the 1991 survey, and the second a permanent deviation after the shock, captured by a dummy variable equal to 1 for observations from 1991 and after, and equal to

1.0 0.8

1989 1990-91 1993-94

0.6 0.4 0.2 0.0 -0.2

0-5

6-11

12-23

24-34

35-46

-0.4 -0.6 -0.8 -1.0

Figure 3. Weight for height in selected years.

47-60

ADJUSTMENT WITH A HUMAN FACE?

0 otherwise. We found a significant one-shot deviation in 1991, with weight for height for kids in the 1991 survey on average 0.203 zscores below other kids (consistent with the results in Table 2). But the existence of a permanent deviation after 1991 was less obvious; the coefficient estimate for the post-shock dummy variable was )0.086 with a t-value of only 1.83. We also performed the same analysis using height for age as the dependent variable. Since height for age is an indicator of chronic or long-term nutritional status, we looked for significant long-term deviations from trend after 1991 and after 1992. We found a significant negative deviation after 1991 of )0.112 z-scores (t ¼ 2:03). This deviation is less pronounced for the post-1992 dummy, where the estimated coefficient was )0.093 and the t-value only 1.61. These results indicate that the shift in height for age occurred in 1992, the year immediately after the shock. 10 Although not conclusive, our initial investigation using raw means, graphs and simple regression suggests that weight for height suffered in 1991. Long-term nutritional status (height for age) of children also appears to be lower after 1991. These simple descriptive regressions tell a different story from that in Behrman and Deolalikar (1991), but this is not surprising for two reasons. First, the data in Behrman and Deolalikar (1991) only go up to 1987, well before the bulk of JamaicaÕs SAP was implemented. Second, their anthropometric data are clinic based, which in Jamaica, tends to underreport the extent of malnutrition compared to survey-based data which we use here (Grosh, Fox, & Jackson, 1991). 5. REGRESSION RESULTS Before displaying results controlling for both age, year and birth cohort, Table 8 in the Appendix A presents information on the number of cohorts and their size. As mentioned earlier, since the SLC collects data on children under age five, the first few rounds of the survey contains information on children who were born before 1989, which is the earliest SLC year that we use. As we would expect, the largest cohorts (given by the last column of numbers in Table 8) are those whose birth year lies within the first few years of the survey rounds, while the smallest cohorts are those born in the last two survey years (1995–96). Given the sample

1133

sizes, we define cohorts according to six-month birth intervals, each interval corresponding to the first six and the last six months of a particular year. Since the 1996 birth cohort is so small (54 observations), we group them with 1995(2) (i.e. kids born in the second half of 1995) and use this as our excluded or base category. (a) Impact of survey year on weight for height We estimate Eqn. (2) with 11 cohort dummy variables as defined above, and seven dummy variables indicating year of survey, along with the usual control variables (sex, 11 parish and regional dummies, and seven age dummies). Summary statistics for all the variables used in the analysis are presented in Table 9 of Appendix A––a subset of these coefficient estimates are provided in Table 4. Note that in a timeseries of cross-sections such as we have here, age, cohort, and survey year cannot be separately identified since each is a linear combination of the other two. In the specifications we present below, identification is achieved through functional form because the age dummies overlap with the year of birth dummies. We have experimented with the alternative proposed by Deaton (1997), where the cohort effects are restricted to sum to zero and have no trend, and found no difference in the estimated coefficients reported below. 12 F -tests at the bottom of Table 4 indicate that the set of year effects (at 1%) but not the cohort effects are significant determinants of the zscore for weight. The year coefficients indicate that even after controlling for previous (or initial) health and economic conditions via cohort effects, mean z-score of children from the 1991 survey is 0.227 z-scores lower than those weighed in 1996, but this difference is not statistically significant. Kids from the 1991 survey are approximately 0.12 standard deviations lighter than those from 1990, but this difference is also not statistically significant (p-value is 0.15); note however that the difference in mean z-score is significantly lower in 1991 compared to 1992 (see bottom of Table 4). (b) Date of measurement effects Children in each round of the SLC are weighed and measured over a period of several months. SLC data have typically been collected at the end of the year, with anthropometric interviews therefore occurring at the end of the

1134

WORLD DEVELOPMENT Table 4. OLS regression results of survey year effects on weight for heighta

Variable 0–2 months (excluded) 3–5 6–8 9–11 12–23 24–35 36–47 48–60 Female child Caregiver years of schooling Household log p  c consumptionb Rural household Year of survey 1989c 1990 1991 1992 1993 1994 1995 1996 (excl.) Observations R2 (%) F P -value cohort effectsd P -value year effects P -value for 1990–91 P -value for 1991–92

Coefficient

T -Statistic

)0.060 )0.259 )0.443 )0.595 )0.731 )0.728 )0.802 0.016 0.014 0.098 0.131

(0.63) (2.65) (4.52) (6.75) (6.83) (5.55) (5.06) (0.61) (2.22) (3.80) (3.93)

0.014 )0.125 )0.227 0.021 0.037 0.151 0.034

(0.06) (0.65) (1.41) (0.17) (0.37) (1.97) (0.50)

6,729 5.16 6.72 0.21 0.00 0.15 0.00

a

13 Parish dummies, date of birth of the child, and a constant, included in the estimation but not reported. Instrumented––see text for details. c Refers to year of survey. d Refers to set of birth date variables. b

year or the beginning of the following year. The timing of the survey, however, was changed to the middle of the year in 1995 so that all anthropometric interviews were done within the same calendar year. We are particularly interested in the outcomes of children weighed around the time of the September 1991 liberalization, which would be children from the 1991 survey. In this survey, 37% of the children in the sample were actually weighed and measured in January or February of 1992. Since child weight is very sensitive to short-run or current living conditions, we test the hypothesis that children in the 1991 survey who were interviewed closer to the shock were more vulnerable and so have lower weight. We therefore refine our year effects by creating dummy variables based on the time within a given year the child was measured,

dividing each year into roughly three periods, corresponding to the beginning, middle and end of the year. For example, the variable 91(3) indicates the child was measured during the end of 1991, while 92(1) indicates the child was measured during the first few months of 1992. For each survey until 1995 we divide children into two groups, typically those weighed and measured at the end of the survey year and those at the beginning of the following year, due to seasonality in consumption associated with Christmas and New Year. We do not separate children in the 1995 and 1996 surveys since these are conducted in the middle of the year when consumption patterns would be fairly stable. The excluded group is children measured in the 1996 SLC. Column 1 of Table 5 presents coefficient estimates for weight for height using these

ADJUSTMENT WITH A HUMAN FACE?

1135

Table 5. OLS estimates of measurement date effects on weight for height z-scorea Variable

Model with rural-measurement date interactions (1)

0–2 months (excluded) 3–5 6–8 9–11 12–23 24–35 36–47 48–60 Female Caregiver years of schooling Household log p  c consumptionb Rural household

(2)

(2a)

Coefficient

T -Statistic

Coefficient

T -Statistic

)0.084 )0.257 )0.433 )0.616 )0.741 )0.754 )0.848 0.009 0.014 0.106

(0.86) (2.57) (4.32) (6.84) (6.75) (5.61) (5.23) (0.34) (2.20) (4.01)

)0.077 )0.256 )0.436 )0.614 )0.741 )0.754 )0.845 0.010 0.014 0.109

(0.78) (2.56) (4.35) (6.81) (6.74) (5.60) (5.22) (0.36) (2.11) (4.07)

0.083

(1.00)

0.120

3.48

Coefficient

Measurement date c

89(3) 90(1) 90(3) 91(1) 91(3) 92(1) 92(3) 93(1) 93(3) 94(1) 94(3) 95(1) 95(2) 96(2) (excluded) Observations R2 F P -value cohort effectsd P -value measurement date effects P -value for 91(1) ¼ 91(3) P -value for 91(3) ¼ 92(1) P -value rural 91(1) ¼ 91(3) P -value rural 91(3) ¼ 92(1) P -value rural measurement dates

T -Statistic

Rural interaction )0.096 )0.054 )0.173 )0.224 )0.342 )0.164 )0.064 0.105 )0.048 0.017 0.155 0.113 0.009

(0.42) (0.24) (0.87) (1.07) (2.02) (0.94) (0.49) (0.80) (0.35) (0.17) (1.87) (1.17) (0.13)

)0.088 )0.133 )0.174 )0.445 )0.572 )0.208 )0.064 )0.091 )0.212 )0.016 0.289 0.049 0.060

0.34 0.57 0.83 1.78 2.89 1.09 0.44 0.58 1.16 0.13 2.58 0.38 0.67

)0.006 0.135 0.001 0.345 0.329 0.110 0.007 0.029 0.279 0.067 )0.219 0.125 )0.105

6,338 5.42 6.10 0.25 0.00

6,338 5.75 5.31 0.26 0.00

0.30 0.07

0.51 0.01 0.95 0.28 0.06

(0.03) (1.22) (0.00) (1.63) (2.16) (0.63) (0.06) (0.22) (1.32) (0.55) (1.70) (0.76) (0.94)

a

13 Parish dummies, date of birth of the child, age of caregiver, and a constant, included in the estimation but not reported. b Instrumented––see text for details. c Refers to date the child was measured. d Refers to set of birth date variables.

measurement dates to capture time effects. These time or ‘‘date’’ effects are jointly significant based on the F -test at the bottom of the

table, although only one individual coefficient is significant. The coefficient estimate for the children measured at the end of 1991 (the

1136

WORLD DEVELOPMENT

variable named 91(3)) is large and negative ()0.342) with a t-value of 2.02, while the estimate for kids measured a few months later (92(1)) is much smaller ()0.164), and these two estimates are statistically different at the 7% level of confidence. Note that average weight for height appears to have begun deteriorating at the end of 1990, when a few financial reforms were initiated and when inflation began its incline (Figure 1). The p-value for statistical difference in effects in 90(1) and 90(3) is 0.09, while the difference in coefficients between 90(1) and 91(3) is highly statistically significant (p ¼ 0:00). The pattern therefore appears to be a decline in average weight beginning at the end of 1990, bottoming out at the end of 1991, and recovery by the end of 1992. Does the difference between 91(3) and 92(1) simply capture differences in consumption patterns due to seasonality of Christmas and New Year? We performed a series of pair-wise t-tests to see whether the coefficients for children from the same survey year were significantly different from each other (for example, 89(3) versus 90(1), 90(3) and 91(1), and so on) for the six survey years where this is possible (89–94). In four of the six cases the p-value of the t-test for differences in coefficients was above 0.50, with the two exceptions being 1991 (the year of the shock) and 1993, where the p-values were 0.07 and 0.01, respectively, but for 1993 there was actually an increase in mean z-score from preChristmas to post (see coefficients in Table 5). 13 These results indicate that seasonal fluctuations in child weight due to Christmas and New Year are not the norm in Jamaica, and thus that the decline in weight observed within the 1991 survey is not simply due to seasonality in consumption associated with Christmas and New Year. Taken at face value, the coefficient estimates imply that children weighed just after the liberalization episode are 0.178 z-scores lighter on average than those weighed a few months later, controlling for age, birth date, sex, and region of residence. (c) Extensions: region, age, income, and education interactions In this section we investigate some further hypotheses to assess the robustness of our results, and to help us better understand the link between the 1991 macro shock and the vulnerability of children. Our earlier review of consumption smoothing mechanisms indicated that urban households were less likely to have

access to traditional informal sources of insurance or smoothing mechanisms, such as extended family or own farm produce. In Jamaica the poverty rate is significantly higher in rural areas, but poor rural households are also likely to have access to domestically produced food such as cassava, yam, and fruits and vegetables, which could have offered a degree of food security during the crisis. To test whether the macroeconomic shock affected urban children differently, we interact the measurement date variables with the rural dummy and report the results in columns 2 and 2a of Table 5. The rural interactions (shown in columns 2a) are jointly significant at the 0.06 level, while the rural interaction for 91(3) is highly statistically significant, the point estimate indicating that during the peak of the crisis the mean weight among rural children was 0.329 z-scores higher than that among urban children. A rural–urban difference, though not as significant, was also present at the beginning of 1991, when the mean z-score was 0.345 higher among rural children (t value 1.63). Figure 4 presents the estimated z-score for urban and rural children according to measurement date; mean weight for height is higher for rural children over the study period, but the difference is very large in 1991 during the macroeconomic crisis, and begins to decline during 1992. 14 As noted above, rural households are poorer than urban ones on average, yet urban children appeared to have suffered more during the 1991 crisis. Access to resources influences nutritional status by increasing potential access to food and quality health care, and is positively associated with nutritional status as estimated in Tables 4 and 5. We tested the hypothesis that richer households were better able to protect children during the crisis by interacting our measure of resources with the date of measurement dummy variables. None of these interaction terms were significant, neither for the full sample nor in urban areas (results available from authors). Motivated by the well-known relationship between childrenÕs health and maternal education, we performed a similar interaction as above, except now using motherÕs (or caregiverÕs) education instead of income. The simple interaction between 91(3) and highest grade attained yielded a positive but insignificant coefficient. We then created dummy variables indicating whether the childÕs caregiver had completed primary, first cycle secondary (grade 9), and second cycle secondary (grade 11), levels of schooling. The coeffi-

ADJUSTMENT WITH A HUMAN FACE?

1137

1.5

1.0 Rural Urban

0.5

0.0

-0.5

-1.0

-1.5

)

(3

89

)

(1

90

)

(3

90

)

(1

91

)

(3

91

)

(1

92

)

(3

92

)

(1

93

)

(3

93

)

(1

94

)

(3

94

)

(1

95

)

(2

95

)

(1

96

Figure 4. Estimated impact of measurement date on weight for height.

cients of these interaction terms were progressively larger (and positive) for higher levels of schooling, and the grade 11 coefficient of 0.116 nearly eliminates the negative effect of 91(3), although it is also not precisely measured (t-value equal to 1.13). 15 Finally, we considered two issues related to the pattern of child growth and development. First, children are more vulnerable during the weaning period which is typically between 6 and 12 months in Jamaica. Second, children who receive a ‘‘shock’’ or nutritional ‘‘insult’’ during a growth spurt may be less likely to recover. We can account for the potential vulnerability during the weaning period by interacting the 1991 time effect with age, to see whether children of specific age groups suffered more in 1991, but we cannot test the second hypothesis because in general children have their growth spurts at different times. When all the age dummies are interacted with a dummy for survey year 1991, none of the interaction terms are individually significant, nor are they jointly significant (p-value of 0.63 for the joint F -test); the same results are found when interacting with kids weighed in November and December of 1991 only (results available from authors). The fact that younger children were not relatively protected during the crisis is probably due to the low rate of exclusive breastfeeding in Jamaica.

(d) Explicit control for macroeconomic conditions The above analysis indicates that children weighed shortly after the liberalization episode are lighter (by 0.178 z-scores) than those weighed just a few months later, and in urban areas the decline in weight was even larger (0.364 z-scores). If this decline in short term nutritional status is due to currency devaluation and the associated price increase of imported food staples, than explicit control for these factors should eliminate or decrease the (negative) impact of the 91(3) date of measurement effect. Using monthly data on the exchange rate and food prices, we construct two variables to measure the potential influence of macroeconomic instability on child weight. The first is the average rate of depreciation of the exchange rate over the six months prior to the date of measurement, and the second is the average rate of food price inflation over the three month period prior to date of measurement (in percent). We experimented with other specifications (different period lengths) and found these two specifications to be the best predictors of child weight. 16 Column 1 in Table 6 presents estimates of the impact of exchange rate depreciation (in percent) over the last six months on the zscore of weight for height. There is a highly

1138

WORLD DEVELOPMENT

Table 6. OLS regressions estimates of exchange rate and food price inflation effects on weight for height z-scorea Exchange rate (1) Caregiver years of schooling Household log p  c consumptionb Exchange rate depreciation last 6 months

0.014 (2.15) 0.118 (4.49) )0.004 (4.12)

(2) 0.014 (2.16) 0.107 (4.04) )0.008 (1.61)

Food price inflation last 3 months Measurement date 89(3)c

(3)

(4)

0.014 (2.17) 0.116 (4.42)

0.014 (2.20) 0.106 (3.99)

)0.022 (5.22)

)0.004 (0.19) )0.091 (0.40) )0.045 (0.20) )0.151 (0.66) )0.201 (0.84) )0.296 (1.00) )0.123 (0.44) )0.061 (0.46) 0.107 (0.81) )0.029 (0.17) 0.034 (0.25) 0.157 (1.88) 0.115 (1.18) 0.015 (0.20)

0.081 (0.32) 0.139 (0.55) )0.003 (0.01) )0.046 (0.20) 0.081 (0.26) 0.293 (0.88) )0.042 (0.32) 0.169 (1.22) 0.231 (1.05) 0.222 (1.37) 0.211 (2.35) 0.171 (1.65) 0.079 (0.99)

90(1) 90(3) 91(1) 91(3) 92(1) 92(3) 93(1) 93(3) 94(1) 94(3) 95(1) 95(2) R2 F -statistic P -value measurement date effectsd P -value 91(1) ¼ 91(3) P -value 91(3) ¼ 92(1)

Food price inflation

4.94 6.82

5.46 5.94 0.00 0.50 0.04

5.10 7.04

5.42 5.90 0.06 0.58 0.09

a

Absolute value of t-statistics in parentheses. 6,338 observations. Also included in the regressions but not reported here are 13 parish and two regional indicators, seven age dummies, sex and date of birth of the child, and a constant. Instrumented––see text for details. c Refers to date the child was weighed. d Joint test for significance of date of measurement variables. b

significant negative relationship between the two, even with the usual set of controls. An additional 1% increase in average exchange rate depreciation over the last six months reduces weight for height by 0.004 z-scores.

When the date of measurement indicators are included in column 2, the 91(3) indicator loses statistical significance relative to the base time period, but the difference between 91(3) and 92(1) is still significant, and the point esti-

ADJUSTMENT WITH A HUMAN FACE?

mate of the difference actually increases to 0.212. Column 3 in Table 6 shows the impact of food price inflation on child weight, and this relationship is also highly negative and statistically significant. An additional 1% increase in average food price inflation over the previous three months reduces the z-score of weight for height by 0.022. But, when date of measurement is included in the model (column 4), food price inflation loses significance, and the difference in estimated (average) z-score between children weighed in late 1991 and early 1992 is slightly reduced to 0.173, significant at the 9% confidence level. In a small open economies such as Jamaica, the value of the exchange rate is a highly political instrument due to the strong link between domestic inflation and currency devaluation. Few studies (if any) have explicitly measured the strength of the relationship between macrovariables such as the exchange or inflation rate, and individual welfare outcomes such as preschooler wasting using micro data. Based on the point estimates in columns (1) and (3) of Table 6, we calculate elasticities of the z-score of weight for height with respect to the exchange rate of )0.36, and )0.86 with respect to food price inflation. During the rapid currency devaluation and inflation of 1991, the (shortterm) elasticities were even higher, reaching )0.45 and )1.24 for the exchange rate and food price inflation, respectively. 17 Both exchange rate depreciation and food price inflation have statistically strong effects on short-term child nutritional status, and according to the results in Table 6, the estimated lower z-scores of weight for height in 91(3) is correlated with food price inflation. To better understand the causal relationship between exchange rate liberalization, inflation, and child health, we performed Granger causality tests (Granger, 1969) to see whether the high food price inflation during this period was due to the rapid currency devaluation observed over the same period. Using lag lengths of four, six, and eight periods, the exchange rate was regressed on past values of itself and the food CPI, and the food CPI was regressed on past values of itself and the exchange rate. We were able to establish one way causality from the exchange rate to the food CPI. 18 But when the same test was implemented in changes, we could not establish one way causality. Past values of the exchange rate were significant predictors of the food CPI,

1139

and past values of the food CPI also significantly predicted the exchange rate, indicating two-way causality. We tried two variations of the model estimated in changes. First, we included lagged values of the money supply as an additional control variable since this could have caused both inflation and exchange rate depreciation. We also estimated switching regressions where all variables were interacted with a dummy variable for the eight month period between August 1991 and March 1992 (when the slide of Jamaican dollar stopped). In neither variations of the model were we able to establish one way causality flowing from the exchange rate to food prices––all models showed dual causality. 19 (e) Long-term effects: height for age The financial liberalization episode had an immediate impact on wages and real incomes, but longer term consequences of this shock on employment and output appear to have been small. As a result, the effect of this episode on longer term (chronic) nutritional status is not likely to have been strong, at least theoretically. Nevertheless, we check to see whether there were any longer term repercussions by analyzing the behavior of height for age among children who were born before or just after September 1991. 20 Specifically, from our working sample we select those children who were born before June 1992. Among these children, those measured before September 1991 were not exposed to the ‘‘treatment,’’ while those measured after that date were exposed to the treatment; for this latter group we define their exposure time as the number of months after September 1991 their measurement occurred. For example, a child born before September 1991 and whose measurement date was September 1992 is defined to have an exposure time of 12 months. Using this sample of children, and nine mutually exclusive exposure time categories, we estimate Eqn. (2) on the z-score of height for age and present the results in Table 7. Almost all of the exposure time variables are negative, indicating that mean height for age is lower among kids who were exposed relative to those who were not. Nevertheless, only one coefficient (13–18 months) is statistically significant. As a whole, it seems like height for age deteriorated significantly about 7–24 months after the shock, and then recovered. The estimated relationship between exposure and height is

1140

WORLD DEVELOPMENT Table 7. OLS estimates of exposure on height for age z-scorea

Variable

(2)

(1) Coefficient

T -Statistic

Variable mean

0–2 months (excluded) 3–5 6–8 9–11 12–23 24–35 36–47 48–60 Female Caregiver years of schooling Household log p  c consumptionb Rural household

)0.492 )0.922 )1.212 )1.246 )1.008 )0.923 )1.049 0.157 0.017 0.272 0.067

(2.66) (4.82) (6.52) (7.56) (5.37) (4.23) (4.20) (4.03) (1.97) (8.12) (1.38)

0.019 0.027 0.027 0.036 0.165 0.202 0.254 0.269 0.493 8.754 8.161 0.579

Exposure in months 0 (excluded)c 1–6 7–12 13–18 19–24 25–30 31–36 37–42 43–58

0.057 )0.192 )0.390 )0.278 )0.028 )0.080 )0.023 )0.039

(0.58) (1.12) (2.64) (1.23) (0.15) (0.32) (0.10) (0.15)

0.368 0.128 0.032 0.206 0.015 0.088 0.019 0.071 0.073

Observations R2 F P -value cohort effectsd P -value for (1–6) ¼ (7–12) P -value for (7–12) ¼ (13–18) P -value for (13–18) ¼ (19–24) P -value for (19–24) ¼ (25–30) P -value for (31–36) ¼ (37–42) P -value for (37–42) ¼ (43–58) P -value for (13–18) ¼ (1–6) P -value for (13–18) ¼ (25–30)

4,638 0.068 8.11 0.00 0.08 0.14 0.20 0.76 0.75 0.57 0.00 0.00

a 13 Parish dummies, date of birth of the child, age of care giver, and a constant, included in the estimation but not reported. b Instrumented––see text for details. c Refers to the length of time after September 1991 the child was measured. d Refers to set of birth date variables.

depicted graphically in Figure 5, and shows a drop in height for age occurring approximately six months after the crisis, and continuing until about 24 months after the crisis, and then returning to the steady state. 6. CONCLUSIONS AND DISCUSSION Using a time series of household surveys we evaluate the consequences for short-term or

current child nutritional status, of JamaicaÕs exchange rate liberalization in September 1991. This economic reform had swift and severe consequences for inflation, real incomes and food prices, and lead to a significant increase in poverty in that year. Our analysis suggests that weight for height declined during this period of adjustment. After controlling for a set of potential intervening variables, we find that the mean weight for height for children 12–23 months of age from

ADJUSTMENT WITH A HUMAN FACE?

1141

2 1.5 1 0.5 0 -0.5

0

6

12

18

24

30

36

42

60

-1 -1.5 -2

Figure 5. Estimated impact of exposure time on height for age.

the 1991 SLC is )0.416, compared to a mean of )0.169 from the previous year (estimated from Table 4) and this difference is statistically significant. When we further distinguish between children from the 1991 survey weighed in November and December of 1991 with those weighed in January and February 1992, we find the former group about 0.178 z-scores lighter than the latter, and the difference significant at the 0.07 level of significance. Moreover, comparison tests between similar groups of children from other survey rounds indicates that the differences in 1991 are not simply due to seasonal consumption patterns brought on by Christmas and New Year. When the exchange rate depreciation is added to the model, the 1991(3) specific fixed effect is still significantly different from 1992(1), but when food price inflation is added to the model this difference disappears. Extensions to our basic model reveal no significant interactions between the 1991 effect and childÕs age or household per capita consumption expenditure. There is, however, an important regional dimension to the impact of the crisis on weight. The impact on urban children was significantly greater (by 0.329 z-scores) than for rural kids, even though the latter come from relatively poorer households. But rural households in Jamaica have more access to own farm produce which seems to have played a mitigating role during this inflationary period. An important result in this study is the strong relationship between the exchange rate, food price elasticity, and childrenÕs current nutritional status. The estimated elasticity of the z-

score of weight for height with respect to food inflation is )0.86; during the crisis period in 1991 this elasticity rose to )1.243. What do our results imply for the broader debate on SAP, pubic policy, and social development? First, due to our unique data base, we are better able than other studies to isolate the impact of currency devaluation on childrenÕs health at the micro level. We find this relationship to be very strong in the context of a small open economy which is dependent on imports of staple foods. Under these circumstances, it appears that adjustment policies which bring about sudden and large changes in the inflation rate can have real affects on childrenÕs short-term nutritional status, especially among urban households who buy their food in the market. Second, due to JamaicaÕs high level of social development, the correlations uncovered here probably serve as a lower bound for similar scenarios in other developing countries. Jamaica has already gone through the epidemiological transition, so that its disease burden and health care problems resemble those of a high-income country such as the United States, rather than a typical poor country. 21 For example, infant mortality is 13 (per 1,000) and the total fertility rate 2.4, compared to means of 41 and 3.0 for all middle-income countries (World Bank, 1997). Maternal education, an important determinant of child health, is also very high in Jamaica, indicating a capacity to process and respond to environmental shocks. 22 Finally, JamaicaÕs maternal and child health care delivery system was recently described by WHO as one of the best in the developing world. For

1142

WORLD DEVELOPMENT

all these reason, the capacity of Jamaican households to respond in a way that would

protect child health is greater than in the typical developing country.

NOTES 1. For example, Handa and King (1997) show that poverty in Jamaica increased dramatically in 1991, the year of major exchange rate reform, but fell very quickly thereafter. They suggest this may have occurred because Jamaica had already been through several initial rounds of market oriented reforms before exchange rate liberalization.

9. They look at deviations from trend in malnutrition in Jamaica using national averages for approximately 11 years (hence their sample has 11 observations). We cover fewer years, but use micro data with over 5,000 observations so that we can control for one of the most important determinants of anthropometric status––age of the child.

2. Given the existence of multiple rounds of Living Standards Measurement Surveys in several developing countries, this technique could also be used elsewhere to assess the social cost of SAP.

10. Results of these descriptive regressions are available from the authors upon request.

3. For a review of JamaicaÕs structural adjustment and reform process, see Lora (1997); Witter and Anderson (1991); Boyd (1987); and Handa and King (1997). 4. Infant mortality is low in Jamaica, 14 per 1,000, and there are no reports of excess mortality during 1990–91. 5. This individual effect includes the initial health endowment of the child as well as past health performance or the stock of health from the previous period. These are some of the effects the cohort specific effect is expected to pick up. 6. As recommended by the World Health Organization, the weight of each child is compared to a reference standard of children of the same sex and height from a well-nourished population (in this case the United States is used as the reference population). The z-score is obtained by subtracting the median weight of children of the same sex and height from the observed or measured weight of children in the sample, and dividing by the standard deviation of weight of the reference population. The resulting variable is measured in standard deviations, as opposed to centimeters or inches. 7. Smaller age groups are not possible due to small cell sizes. 8. The rate of wasting, defined as the proportion of children with weight for height z-score less than or equal to )2, also increased significantly in 1990 and 1991. The average rate of wasting over the whole period was 3.9%, compared to 6.1% and 5.3% in 1990 and 1991, respectively. In 1992, the year after the shock, the rate of wasting was 2.8%.

11. There are no significant differences by sex, a common finding in the English-speaking Caribbean, where female headship and labor force participation is very high, and girls routinely outperform boys in school. 12. Deaton (1997) suggests restricting the year effects to zero in order to identify age and cohort effects. But since we are precisely interested in estimating year effects, and since these effects are unlikely to sum to zero in this particular case, we restrict the cohort effects instead. Note that the cohort effect is unlikely to affect short-term nutritional status. 13. These results are available from the authors upon the request. 14. Rates of wasting were also significantly higher in urban areas, averaging 8% in 1990–91 compared to only 4% in rural areas during the same time period. 15. On average 29% of female caregivers have completed grade 11 or higher in the sample. But, the interaction between this variable and the variable 91(3) yields only 170 positive observations out of a total sample size of 6,331, or 2.7%, which also explain the low t-statistic. 16. For exchange rate depreciation the last three periods average was also statistically significant but not as strong quantitatively. Similarly, for food price inflation the last six periods average was also statistically significant but quantitatively not as large. 17. These estimates were derived from a model interacting the exchange rate and food inflation (separately) with the variable 91(3).

ADJUSTMENT WITH A HUMAN FACE? 18. In other words, past values of the exchange rate were significant predictors of the food CPI, but past values of the food CPI were not significant predictors of the exchange rate. 19. Results of the Granger causality tests are available from the authors upon request. 20. Children born well after that date were not exposed to the crisis during their life.

1143

21. JamaicaÕs health care burden comes from curative and chronic illnesses rather than infectious and childhood diseases. See Handa and Neitzert (1998) for a detailed discussion.

22. Female illiteracy and secondary enrollment is 11 and 70%, respectively, compared to averages of 23 and 61% for all middle income countries (World Bank, 1997).

REFERENCES Alderman, H., & Garcia, M. (1994). Health security and food security: explaining the levels of nutritional status in Pakistan. Economic Development and Cultural Change, 42(3), 485–507. Alleyne, D. (2001). The dynamics of growth, employment, and economic reforms in Jamaica. Social and Economic Studies, 50(1), 55–126. Baker, M., & Benjamin, D. (1994). The performance of immigrants in the Canadian labor market. Journal of Labor Economics, 12(3), 369–405. Behrman, J. (1988). Intra-household allocation of nutrients in rural India: are boys favored? Do parents exhibit inequality aversion? Oxford Economic Papers, 40, 32–54. Behrman, J., & Deolalikar, A. (1988). Health and nutrition. In H. Chenery, & T. N. Srinivasan (Eds.), Handbook of development economics (Vol. 1). Amsterdam: North-Holland. Behrman, J., & Deolalikar, A. (1991). The poor and social sectors during a period of macroeconomic adjustment: empirical evidence from Jamaica. The World Bank Economic Review, 5(2), 291–313. Berry, A. (1995). The social challenge of the new economic era in Latin America. FOCAL discussion paper no. 1995-8. Centre for International Studies, University of Toronto. Boyd, D. (1987). The impact of adjustment policies on vulnerable groups: the case of Jamaica, 1973–1985. In G. Cornea, R. Jolly, & F. Steward (Eds.), Adjustment with a human face (Vol. 2). New York: UNICEF. Deaton, A. (1997). The analysis of household surveys: a microeconometric approach to development policy. Baltimore: The Johns Hopkins University Press. Economic and Social Survey of Jamaica (various years). Kingston, Jamaica: The Planning Institute of Jamaica. Foster, A. (1995). Prices, credit markets and child growth in low-income rural areas. Economic Journal, 105, 551–570. Granger, C. W. J. (1969). Investigating causal relations by econometric models and cross-spectral methods. Econometrica, 37, 424–438. Grosh, M. E., Fox, K., & Jackson, M. (1991). An observation on the bias in clinic-based estimates of malnutrition rates. Policy, research, and external

affairs working paper no. WPS 649. Washington, DC: The World Bank. Handa, S. (1999). Maternal education and child height. Economic Development and Cultural Change, 47(2). Handa, S., & King, D. (1997). Structural adjustment policies, income distribution and poverty: a review of the Jamaican experience. World Development, 25(6), 915–930. Handa, S., & Neitzert, M. (1998). Chronic illness & retirement in Jamaica. LSMS working paper no. 131. Washington, DC: The World Bank. Jacoby, H., & Skoufias, E. (1997). Risk, financial markets, and human capital in a developing country. Review of Economic Studies, 64, 311–335. Jolly, R., Stewart, F., & Cornia, G. (Eds.) (1984). Adjustment with a human face. UNICEF, New York. Lora, E., 1997. A decade of structural reforms in Latin America: what has been reformed and how to measure it. In Presented at the annual meeting of the Inter-American Development Bank, Barcelona. McKinnon, R. (1973). Money and capital in economic development. Washington, DC: Brookings Institutions. Morduch, J. (1995). Income smoothing and consumption smoothing. Journal of Economic Perspectives, 9, 103–114. Paxson, C. (1992). Using weather variability to estimate the response of savings to variable income in Thailand. American Economic Review, 82, 15–33. Shaw, E. S. (1973). Financial deepening in economic development. New York: Oxford University Press. STATIN (various years). The labour force survey. Kingston, Jamaica: The Statistical Institute of Jamaica. Strauss, J. (1990). Households, communities, and preschool childrenÕs nutrition status: evidence from rural C^ ote dÕIvoire. Economic Development and Cultural Change, 9, 231–262. Strauss, J., & Thomas, D. (1995). Empirical modeling of household and family decisions. In J. Behrman, & T. N. Srinivasan (Eds.), Handbook of development economics (Vol. 3A). Amsterdam: North-Holland. Townsend, R. M. (1995). Consumption insurance: an evaluation of risk-bearing systems in low income economies. Journal of Economic Perspectives, 9, 83– 102.

1144

WORLD DEVELOPMENT

Witter, M., & Anderson, P. (1991). The distribution of the social cost of JamaicaÕs structural adjustment 1977–1989, Mimeo. University of the West Indies. World Bank (1997). World development report. New York: Oxford University Press.

APPENDIX A See Tables 8 and 9 on facing page.

ADJUSTMENT WITH A HUMAN FACE?

1145

Table 8. Number of observations in each birth cohort by survey year Birth year

Year of survey 1989

1990

1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996

222 226 236 234 223 5

7 126 141 141 151 102

Total

1,146

668

Total

1991

1992

1993

10 124 136 136 129 87

1 27 293 357 316 293 198 4

6 124 149 144 149 154

622

1,489

726

1994

7 159 135 153 172 113

739

1995

80 148 155 164 165 57 769

1996

78 115 176 171 135 54

229 363 528 810 998 940 885 770 670 449 192 54

729

6,888

Table 9. Full sample summary statistics for control variables Variable

Mean

S.D.

Parish dummies

Mean

S.D.

Kingston St. Andrew St. Thomas Portland St. Mary St. Ann Trelawny St. James

0.034 0.150 0.027 0.035 0.052 0.061 0.044 0.076

(0.18) (0.36) (0.16) (0.18) (0.22) (0.24) (0.20) (0.27)

Hanover Westmoreland St. Elizabeth Manchester Clarendon St. Catherine

0.029 0.067 0.078 0.066 0.116 0.166

(0.17) (0.25) (0.27) (0.25) (0.32) (0.37)

Female Rural Log p  c expenditure Weight for height z-score CaregiverÕs schooling CaregiverÕs age Height for age z-score

0.493 0.572 8.153 )0.102 8.681 37.459 )0.076

(0.49) (0.49) (0.58) (1.10) (2.52) (15.65) (1.26)

Age (in months) dummies 0–2 3–5 6–8 9–11 12–23 24–35 36–47 48–60

0.033 0.043 0.045 0.048 0.214 0.209 0.211 0.196

(0.18) (0.20) (0.21) (0.21) (0.41) (0.41) (0.41) (0.40)

Year dummies 1989 1990 1991 1992 1993 1994 1995 1996

CPI 0.166 0.097 0.090 0.217 0.105 0.107 0.111 0.106

(0.37) (0.30) (0.29) (0.41) (0.31) (0.31) (0.31) (0.31)

1989 1990 1991 1992 1993 1994 1995 1996

1.00 1.21 1.60 3.20 3.85 5.32 6.40 8.40