Fertility correlates in China

DISCUSSION PAPER Report No. DRD 99 FERTILITY CORRELATES IN CHINA Jacques van der Gaag August 1984 Development Research Department Economics and Research Staff World Bank The World Bank does not accept responsibility for the views expressed herein which are those of the author(s) and should not be attributed to the World Bank or to its affiliated organizations. The findings, interpretations, and conclusions are the results of research supported by the Bank; they do not necessarily represent official policy of the Bank. The designations employed, the presentation of material, and any maps used in this document are solely for the convenience of the reader and do not imply the expression of any opinion whatsoever on the part of the World Bank or its affiliates concerning the legal status of any country, territory, city, area, or of its authorities, or concerning the delimitations of its boundaries, or national affiliation. FERTILITY CORRELATES IN CHINA Jacques van der Gaag Development Research Department The World Bank August 1984 Excellent research assistance by Manon Muller is gratefully acknowledged. Special thanks go also to Wendy Shinn for expert typing and editorial assistance. * Mr. van der Gaag is a staff member of the World Bank. The World Bank does not accept responsibility for the views expressed herein which are those of the author and should not be attributed to the World Bank or to its affiliated organizations. The findings, interpretations, and conclusions are the results of research supported by the Bank; they do not necessarily represent official policy of the Bank. The designations employed, the presentation of material, and any maps used in this document are solely for the convenience of the reader and do not imply the expression of any opinion whatsoever on the part of the World Bank or its affiliates concerning the legal status of any country, territory, city, area, or of its authorities, or concerning the delimitation of its boundaries, or national affiliation. FERTILITY CORRELATES IN CHINA Table of Contents Page Abstract ii I. Introduction 1 II. The Data 6 III. Estimation Results 11 IV. Rates of Change in the Birthrate 15 V. Relative Rates of Change 19 VI. Conclusion 24 - 11 - ABSTRACT The paper's main objective is to analyze the relationship between fertility and economic well being in the People's Republic of China. We employ a rather unique data set consisting of three repeated cross-sections for the years 1975, 1980 and 1982 over 46 counties in 4 provinces. During our search for this relationship we show (1) how easy it is to obtain "the standard" but misleading results from simple regression analysis, and (2), that even rather stable income-fertility profiles do not yield sufficient information to predict the direction, let alone the magnitude, of change of fertility patterns in China. The first problem is due to the highly heterogeneous environments in which the data were collected (especially large inter-provincial differ- ences). The latter stems from the strong government population policies in China, which have created a disequilibrium situation in which discrete changes in policy, rather than marginal changes in economic variables, are the dominant factors influencing fertility. The paper should serve as a warning against the use of standard cross-section analysis for prediction purposes in such an unstable situation. -.1.- FERTILITY CORRELATES IN CHINA I. Introduction China's remarkable success in reducing fertility is well known. The crude birth rate declined more than 50 percent during the period 1965-1982, a record unmatched in the developing world.i/ As a result, China's current birthrate of 21.1 (1982) is much lower than one would expect from a country at this level of development (in 1981 China's estimated per capita rural income was 223Y in 1982 or approximately $110),2/ Some recent studies have attempted to address the question of to what extent the decline of China's birthrate follows the "normal" pattern in developing countries and to what extent China's achievements are the direct result of various China specific population control measures taken by the government. The "normal" pattern of fertility in developing countries suggest that socioeconomic development can be expected to be accompanied by a decline in birthrates. To quote the first conclusion of Birdsall and Jamison (1983), "Differences in level of development across regions of China are associated with differences in fertility levels. China follows the general pattern: fertility is lower in high-income regions.," The Birdsall/Jamison result stems from a cross-section regression analyses of province level data. A similar analyses was conducted by van der Gaag (1983), using commune level data, He too found a negative income gradient for the birthrate, in a simple birthrate/income regression. However, 1/ World Development Report, 1984. 2/ Statistical Yearbook of China, 1981. -2- a somewhat more elaborate analyses of the data, including information on the proportion of birth control users and one-child certificate holders, showed less straightforward results. Birth control programs and especially the economic disincentives for having more than one child were found to be very effective. Consequently, "further economic development ... will reduce the effectiveness of the current population policy" (op cit. p. 31) and the overall effect of socioeconomic development on the birthrate in China becomes ambiguous. In this paper we will supplement the evidence from province and commune level data by an analysis of population data collected at the country level. Data from 46 counties in 4 provinces are available for the years 1975, 1980 and 1982. Without further discussion of these data we first present, for 1982, the estimation results of a linear regression equation that explains across-county variation in the birthrate by variation in per capita income and the proportion of illiterates in the population (Table 1). Income is estimated as per capita agricultural and per capita industrial production. The former is likely to be the better proxy. The significant coefficient implies an income elasticity of about -0.34. Birdsall and Jamison report a comparable elasticity of -0.40, van der Gaag reports -0.59. Similar results as those reported in Table 1 were obtained for 1975 and 1980 (not shown). All results regarding the income effects seem to overwhelmingly support the general idea that socioeconomic development reduces fertility. This conclusion is being enhanced by the significant positive effect of the illiteracy rate on the birthrate in Table 1. -3- Table 1: Regression Results; Birthrate in China, 1982 (standard errors in parentheses) Agricultural Industrial Production Production Illiteracy Per Capita Per Capita Rate Constant -0.021 -0,0002 0.339 15.02 0.365 (0.007) (0.0007) (0.093) Of course many variables other than income and literacy are important determinants of fertil!,ty. And, in fact, it is quite possible that factors correlated with income, rather than income per se are behind the findings presented above. Consequently, one should be very careful in interpreting results based on such a simple regression analysis. As we will see below, the estimation results turn out to be quite misleading. As stated above, the county level data stem from 46 counties situated in 4 provinces. The provinces are Heilongjiang, Shandong, Sichuan and Ningxia. The implicit assumption that we are dealing with a homogenous sample is clearly violated for the five counties situated in Ningxia. First of all, for 1982, the average birthrate in Ningxia is 28.32 as compared to an average of 17.45 for all 46 counties. But, more importantly, Ningxia also differs in various other aspects relevant to population growth and policy. The main issue is Ningxia's large minority population (53.8 percent, compared to 7.3 percent for the entire sample). The Chinese population policy for minorities is very different from its overall policy: minorities are explicitly exempted from all Chinese -4- attempts to reduce population growth. The counties in Ningxia also have the highest proportion of the population living in rural areas, they provide the highest levels of MCH services per capita and have a population age distribution far different from that of the counties in the other provinces. Moreover, in this sample, all these factors are highly correlated with both per capita income and the literacy rate, if only because these last two variables are much lower in Ningxia than in the other provinces. One way of "controlling" for all these variables (and other province specific factors) is to include dummy variables for the provinces involved. The results are presented in Table 2. Province 1 is Ningxia. Table 2: The Birthrate; Regression Results, 1982 Including Province Dummy Variables (standard error in parentheses) Agricultural Industrial Province Production Production Illiteracy Per Capita Per Capita Rate 1 2 3 Constant R -0.010 -0.003 -0.078 11.342 2.185 1.117 16.79 0.628 (0.006) (0.005) (0.123) (2.04) (1.83) (1.358) The province effect dominates the regression. The income effect, though still negative, becomes statistically insignificant. The "normal" positive and very significant effect of the illiteracy rate, as shown in Table 1, becomes negative and insignificant. The provinces numbered 2 and 3 do not seem to differ significantly from the deleted one. The only conclusion we can draw from Table 2 is that the birthrate in the Ningxia counties is higher than -5- in the other counties. Unfortunately, it seems now fair to say that this should also be the only conclusion to be drawn from the results presented in Table 1. Fortunately the data allow us to do better than to control for inter- province differences by using dummy variables only. Some of the variables mentioned above are available for each county in our sample. In addition, data are available for 1975, 1980 and 1982, so that province specific charac- teristics that do not change much over time can be "controlled for" by looking at rates of change in the birthrate, rather than at levels. Given the results in Tables 1 and 2, however, it seems prudent to "homogenize" the sample by restricting ourselves to the 41 counties situated in Shandong, Heilongjiang and Sichuan only. Ningxia differs too much from the other provinces to expect similar patterns in fertility correlates 1/ In the next two sections we will discuss the data and regression results regarding variation in the levels of the birthrate across counties. In Sections IV and V we will discuss absolute and relative levels of change, respectively. Section VI concludes. 1/ We would be somewhat more inclined to leave the Ningxia data in the sample, if birth rates for minorities and the "majority" (i.e., Han) population would be available separately. -6- II. The Data The Birthrate The data have been collected during 1983 for each of the 46 counties that participated in the first World Bank project on health and medical education in China.1/ Information is available for 1975, 1980 and 1982. Table 3 shows summary statistics for the crude birthrate. Table 3: Crude Birthrate, 41 Counties, 1975, 1980, 1982 Standard Mean Deviation 1975 23.50 6.07 1980 12.71 2.87 1982 16.06 2.31 The 1975 average is almost exactly the same as the national average of 23.13 (e.g., King, 1983, Table 1). The period 1975-1980 saw a dramatic change in the birthrate in these counties; the 1982 rate of 12.71 births per 1,000 population not only cuts the 1975 rate almost in half, it is also much lower than the estimated national average of 21.1. The drop, indeed, is so large that we may have to question the reliability of the data. At the very least, the discrepancy between this sample average and the national average should warn us against making unconditional national generalizations. Just as revealed by national data, the period 1980-1982 saw an increase in the birthrate in this sample from 12.71 to 16.06. 1/ From here on we will delete all information on the five Ningxia counties. So the analysis refers to 41 counties only; 16 in Sichuan, 15 in Shandong and 10 in Heilongjiang. -7- King (1983) gives various explanations for this increase. First of all, the 1980 Marriage Law, which had the effect of voiding local regulations regarding the minimum age of marriage, may have led to a significant increase in the number of marriages. A second major cause may have been the introduction of the responsibility system. The responsibility system has led to a considerable increase in the living standards of the peasants. And, as suggested in the Introduction, if the economic disincentives of the one-child family policy are indeed mostly responsible for its relative success, an increase in income will make it more affordable to have the second (and third) child. Thus resulting in an increase of the birthrate. This explanation, of course, presumes that "children" are a normal good. It also contradicts the "normal" pattern of declining fertility rates with increasing development. It should be noted here that this "normal" pattern in developing countries refers to fertility levels as high as 40 in the Middle East and North Africa, and even close to 50 in sub-Saharan Africa. The pattern is being explained, first of all, by very high infant mortality rates: in some parts of Africa one out of five children die before reaching the age of one, in India, Pakistan and Bangladesh the number is one out of seven.1/ Secondly the "normal" patterns are being explained by the fact that an increase in development, as reflected in an increase in (potential) income, increases the opportunity cost of children, and thus reduces their demand. And finally improved economic status reduces the need for children as old age insurance for their parents. 1/ World Development Report, 1984. -8- Some of these factors, for instance the "social security" explanation, may still play a role in China. However, China's birthrate is already about 50 percent below the level that can be expected for a country of its stage of development, while the infant mortality rate is only 45 per thousand.1! The unusually low birthrate is mainly the result of a very active government population policy. With a crude birthrate already in the neighbor- hood of 20 per 1000, the question is whether a further increase in income will be accompanied by a further reduction in fertility or whether it will result in an increase in the number of children per family. The recent 1980-82 increase in the birthrate suggests that the latter development may be the most likely outcome. In relative terms, a significant increase in the overall birthrate can already be observed if a small proportion of the one-child families have a second child, which is equivalent to a 100 percent increase in the number of children. Again this situation differs very much from a decrease, say from sevenl to six chil ceL, wnich is "only" a fifteen percent change. In sum, since the sixties, China has seen a dramatic decline in the birthrate. Its current low level is unique among countries at China's level of development. This unique low level alone makes it worthwhile to inves- tigate the fertility/income relationship using China specific data, rather than to rely on evidence from other countries. 1/ The Health Sector in China. -9- Income and Other Exogenous Variables As stated above, data used in this study stem from counties situated in these provinces. Various measures of per capita income are available, but none can be considered ideal. For each of the three years we have total agricultural and total industrial output. The former is generally considered to be the better proxy for a county's average economic welfare. The latter often reflects the activities of state- or province-run enterprises that, other than by providing employment, do not directly contribute to the wealth of the county. For 1982 the data also give the county's per capita distributed income. Prior to the introduction of the responsibility system, this measure would have been preferable over the other ones. However, in 1982, three years after the introduction of the responsibility system, distributed income is just one component of total income, and not necessarily the most important one. Table 4 gives the summary statistics of all income variables. We see a steady increase in income, both for the period 1975-1980 and for 1980-1981. Table 4: Income Variables, Yuan Per Capita 1975 1980 1982 Standard Standard Standard Mean Deviation Mean Deviation Mean Deviation Per Capita Agricultural 175 58 229 67 297 91 Output Industrial 91 60 157 103 185 1l1 Output Distributed -- -- -- -- 184 184 Income - 10 - Various other variables that are expected to have an impact on the birthrate are available. First of all, we include in our subsequent analyses the proportion of the population living in rural areas. The average birthrate in rural China was 17.9 in 1979. In urban areas it was only 13.9 X/ We expect counties with a relatively large urban population to have a relatively low birthrate. A large proportion of all birth control activities is carried out by the staff of Maternal and Child Health Centers (MCHs). Therefore, we will include the MCH-staff/population ratio in our subsequent analyses. A well developed MCH center may imply a lot of birth control activities. It may also be the result of a large demand for post-partum and child health care activities. This makes its ultimate effect on the birthrate, in a cross- section analysis, ambiguous. Three variables are available for 1982 only: illiteracy rates, the size of the minority population and the proportion of the population between 16 and 50 years of age, The first two variables are expected to have a positive impact on the birthrate. We will use the third variable as a rough proxy of the proportion of married women of reproductive age. Table 5 presents summary statistics. 1/ The Health Sector Report in China. - 11 - Table 5: Summary Statistics 1975 1980 1982 Standard Standard Standard Mean Deviation Mean Deviation Mean Deviation Proportion 0.895 0.08 0.889 0.08 0.888 0.08 Rural Population MCH-staff per 0.019 0.02 0.028 0.02 0.035 0.02 100 Population Illiteracy Rates, 23.3 6.2 People 12 Years and Over Proportion 1.59 2.87 Minorities Proportion Between 50.25 3.37 16 and 50 years of age III. Estimation Results In all that follows we will use a simple linear regression model to analyze the data. In this section the birthrate in 1975, 1980 and 1982 is the dependent variable. Table 6 presents the estimation results. In order to investigate non-linearities in the fertility/income relationship we include the square of per capita agricultural and per capita industrial production to the equation. In addition to the available explanatory variables we also added two dummy variables representing Heilongjiang and Shandong province, respectively. Sichuan is the excluded province. The dummy variables represent province specific factors not included in our data. Table 6: Regression Results; Crude Birthrates, 1975, 1980, 1982. X:X Per Capita Per Capita Proportion Proportion Agricultural Industrial 2 Rural MCH-staff/ Proportion Proportion 16-50 Years Production Production* X1 22 Population Population Illiterate Minorities of Age Heilongjiang Shandong C R2 1975 0.009 -0.116 -0.056 0.305 -4.767 -51.680 -0.867 -4.019 37.694 0.340 (0.066) (0.065) (0.157) (0.236) (16.422) (58.112) (3.268) (2.272) 1980 -0.012 0.056 0.019 -0.105 0.067 17.317 2.362 2.397 7.286 0.586 (0.33) (0.016) (0.067) (0.030) (6.594) (21.623) (1.085 (0.874) 1982 0.059 0.008 -0.098 -0.016 2.341 28.889 -0.238 1.697 3.519 0.171 (0.024) (0.014) (0.030) (0.020) (8.379) (23.523) (1.194) (1.057) 1982 0.062 0.006 -0.103 -0.018 5.408 26.544 -0.114 -0.213 -0.025 -1.214 2.472 1.930 0.178 (0.025) (0.014) (0.040) (0.020) (9.077) (25.392) (0.095) (0.174) (0.114) (1.370) (1.383) /* The coefficient has been multiplied by 1,000 for ease of presentation. - 13 - For 1982 we estimated the equation with and without the additional information on illiteracy rates, minorities and the proportion of the population between 16 and 50 years of age. None of these variables showed a significant influence. This is probably due to the fact that the first two do not show much variance in the sample, while the latter is too rough of a proxy for the proportion of married women in the reproductive age. Important for the interpretation of the results for 1975 and 1980 is the fact that excluding these variables from the 1982 data does not seem to influence the results for other variables significantly. For 1975 the results are quite disappointing. A simple birthrate/ income regression (not shown ip Table 6) yields a coefficient of -0.028 (standard error 0.014) for per capita agricultural production, and -0.047 (0.014) for per capita industrial production. These results again suggest the "usual" pattern of declining fertility with economic development. The results of the expanded regression presented in Table 6, are quite different. The equations include the squared income variables, rural population, MCH-staff and province dummies as independent variables. The only statistically significant results show that the average county in Shandong has a crude birthrate four points below that of a comparable county in Sichuan (T-value 1.82). All other coefficients are not significantly different from zero. The 1980 results are somewhat more interesting. Both industrial and squared industrial production turn out to have a significant impact. The fact that industrial rather than agricultural production is related to the birthrate comes as a surprise. The results imply an inverted U-shaped relationship between the birthrate and per capita industrial production. The maximum birthrate is reached at about 265M which puts most of the observations in the current sample on the rising side of the curve. ER-024/JVD/8.17.84 - 14 - The birthrate in Shandong now seems to exceed the one in Sichuan. The same seems true for Hiilongjiang. The other variables do not show a significant effect. In 1982 no inter-provincial differences (as measured by the dummy variables) exist. The effect of industrial output on the birthrate shows very much the same pattern as in 1980, but is no longer statistically significant. Agricultural production, however, now has a very significant impact. The results are very similar to those for industrial production in 1980: an inverted U-shaped relation, with a maximum at 300Y. So most countries are again on the rising side of the curve. As in all cases discussed above, none of the other variables show a significant impact. In sum, the results show a very unstable structure. The province dummy variables suggest that there are unknown, province specific, develop- ments that strongly influence fertility. So much so that, conditional upon the other variables, the fertility ranking of the three provinces can change from period to period. The income variables show no relationship at all in 1975, but some indication of a positive relationship (over the range of the observations) both in 1980 and 1982. But even this remotely stable result has to be interpreted carefully since it is obtained with two different income proxies: industrial per capita production in 1980 and agricultural per capita production in 1982. The general impression of instability undermines one of the implicit assumptions needed to base projections on the results of this type of cross- section analysis: the assumption of a relatively balanced (equilibrated) situation. In a balanced situation the simple cross-section results can be - 15 - used (carefully) to make projections on the basis of scenarios regarding the development of the independent variables. If, on the other hand, the structure is plagued by unobserved structural shocks, which need time to adjust to, the descriptive regression, equations are just that: descriptions of statistical relations within the sample. Any attempt to make projections on the basis of these regressions is futile,, The data seem to indicate that we are in this unfortunate situation. The regressions on the differences in the level of the birthrate do not seem to give us much information about the likely magnitude or even direction of future changes. In fact, they yield the "wrong" information. For instance, given the estimation results for 1975 and the increase in income over the period 1975-1980, we would have predicted an increase in the crude birthrate, while, as we saw above, it decreased almost 50 percent. Since it is our ultimate goal to increase our ability to make plausible projections on the development of fertility in China, in the next sections we will employ various models that focus on changes in the birthrate, rather than on levels of fertility. IV. Rates of Change in the Birthrate In this sample, between 1975 and 1980, the crude birthrate fell almost 11 points, from 23.50 to 12.71. From 1980 to 1982 it went up, on average, more than 3 points, from 12.71 to 16.06. In this section these changes over time, rather than the average levels, will be .used as the dependent variable. We first investigate the change in the birthrate from 1975 to 1980. We estimate to what extent the 1975 level of variables can be used as a predictor for the (absolute) change in the crude birthrate during the - 16 - following five years. Then we use, in the same way, the 1980 information on the independent variables to predict the change in the crude birthrate from 1980 to 1982. Thus, in all cases, the change in the birthrate over the period observed is being explained by information collected at the beginning of the period. The results are presented in Table 7. We estimated the regression equations with and without the beginning- of-the-period birthrate level as an explanatory variable. As the Table shows, the beginning of the period level is a very important predictor for future rates of change. The results for 1975-1980 shows that the absolute decrease in the birthrate is larger in areas with a relatively high birthrate at the beginning of the period. Similarly, for 1980-1982, the average increase is smaller. Simply stated relative changes are small. Which factors, other than the initial level, are associated with the observed changes in the birthrate? The regressions presented in the second and fourth row of Table 7 show a very stable pattern for the income/change-in- fertility relationship, with income measured by per capita agricultural production. Both regressions indicate an inverted U-shaped relationship with a maximum expected change in birthrate around 225Y (first period 215Y, second period 239Y). For the first period this implies that most counties are on the rising side of the curve, i.e., in most cases higher income counties saw a smaller decrease in the birthrate than lower income counties. For the second period, the counties are evenly spread around the maximum. Table 7: Regression Results; Absolute Change in the Crude Birthrate; 1975-1980 and 1980-1982 (standard error in parentheses) 1 22 MCH Beginning X 2 1,000 1,000 Rural Staff Rate Heilongjiang Shandong C R 1975- 0.055 0.173 -0.094 -0.493 -0.019 54.558 2.906 6.549 -29.566 0.585 1980 (0.068) (0.066) (0.159) (0.238) (0.166) (58.717) (3.306) (2.313) 1975- 0.061 0.065 -0.142 -0.209 -0.063 6.116 -0.927 2.172 2.713 5.670 0.941 1980 (0.025) (0.026) (0.060) (0.032) (0.062) (22.327) (0.067) (1.243) (0.912) 1980- 0.105 -0.051 -0.220 0.097 0.027 4.681 -2.462 -1.096 -5.069 0.405 1982 (0.043) (0.021) (0.083) (0.040) (0.088) (28.743) (1.442) (1.169) 1980- 0.097 -0.016 -0.208 0.031 0.027 15.615 -0.630 -0.975 0.408 -0.457 0.528 1982 (0.039) (0.022) (0.080) (0.042) (0.078) (25.867) (0.209) (1.376) (1.156) - 18 - For the period 1975-1980, the beginning of the period per capita level of industrial production also shows a significant effect. Again an inverted U-shape, with the maximum at 156Y, which exceeds the 1975 level of industrial production for most counties. For 1980-1982 this effect is no longer statistically significant. The results in Table 7 further show that the large decrease in the birthrate during the period 1975-1980 was, ceteris paribus, smaller in counties with a large rural population; in these counties, the average increase during the period 1980-1982 was also somewhat smaller. Both results, however, are statistically insignificant, as are the results regarding the MCH-staff. For 1975-1980, differences among the provinces are statistically significant: both Heilongjiang and Shandong saw a smaller reduction in the birthrate than Sichuan, holding all other variables constant. For 1980-1982 all provinces seem to behave similarly. In sum, we find significant relationships between indicators of development at the beginning of the period, and the absolute change of the birthrate during the period. The strongest predictor regarding these changes is the beginning of the period level of the birthrate. The results suggest that two further improvements can be made: First, a relative change model may be more appropriate than an absolute change model. Secondly, for policy purposes it may be more useful to have a model that relates "changes to changes". In the next section we will show to what extent these alternative specifications change our results. - 19- V. Relative Rates of Change Table 8 shows estimation results of a model in which the relative rate of change has been regressed on the beginning of the period levels of the independent variables. For 1975-1980 the average birthrate dropped 41 percent, for 1980-1982 it increased 32 percent. Again two versions have been estimated. One that excludes and one that includes the beginning of the period level of the birthrate as an independent variable. And again this variable shows a very significant impact. From Table 7 we could conclude that absolute changes are smaller for counties close to the mean. From Table 8 we draw an even stronger conclusion: the relative change for the period 1975-1980 -- negative on average -- is larger for counties starting at a high birthrate level. The relative change for the period 1980-1982 -- positive on average -- is smaller for counties starting at a high level. This observation may be of considerable importance since it is consistent with the hypothesis that population policy is both endogenous and effective. If, indeed, in high fertility areas stronger population policy measures are being taken, while in already low fertility areas some slippage is allowed to occur, these results can be expected. We should emphasize here that the effect of "population policy" is outside our model. In fact, most of the "action" regarding the birthrate seems to be outside of our model. The question asked by our models is: can information on the beginning of the period level of the birthrate and on the beginning of the period level of economic and other variables, say anything about future changes in the birthrate. The answer, both from Table 7 and R-148-T8 Table 8: Regression RestIts; Relative Change in the Crude Birthrate; 1975-1980 and 1980-1982 (standard error in parentheses) X2 2 MCH Beginning Heilongjiang Shandong X X 1,000 1,000 Rural Staff Birthrate D D C R2 1975- 0.206 0.481 -0.394 -1.373 -0.008 203.654 8.369 16.725 -103.751 0.668 1 1980 (0.163) (0.158) (0.386) (0.576) (0.401) (141.389) (7.995) (5.593) 1975- 0.216 0.300 -0.474 -0.893 -0.082 127.749 -1.567 7.127 10.239 -43.676 0.802 I 1980 (0.126) (0.128) (0.238) (0.456) (0.310) (111.107) (0.335) (6.184) (4.538) 1980- 0.603 -0.441 -1.380 0.854 0.413 -57.423 -24.110 -21.668 -10.153 0.431 1982 (0.426) (0.203) (0.877) (0.389) (0.862) (282.540) (14.121) (11.434) 1980- 0.515 -0.032 -1.236 0.083 0.416 70.507 -7.367 -6.722 -4.07 43.817 0.605 1982 (0.355) (0.200) (0.732) (0.382) (0.718) (237.746) (1.924) (12.649) (10.621) - 21 - Table 8 is unambiguously positive. Counties with a birthrate far from the mean at the beginning of the period tend to change faster (to the mean) than counties that started off closer to the mean. The income profile, for both absolute and relative changes, is very stable: for both periods, counties with an agricultural production around 225Y per capita show the smallest decrease (1975-1980) or the largest increase (1980-1982). Holding all other variables constant, Heilongjiang and Shandong show a higher absolute change and a higher relative change than Sichuan, for 1975-1980. On the other hand, all three provinces "look alike" for 1980-1982. Thus the models seems to reveal in which counties changes will be largest once changes in the birthrate take place. The models fail to predict, however, the direction of the major changes, i.e., the decrease during 1975- 1980 and the upswing during 1980-1982. Of course the regression results "reveal" these trends in the constant terms, which change, in Table 8, from -43.7 to +43.8, indicating a major structural change. The point is that major structural changes rather than marginal differences in the independent variables seem to drive the system. In other words, most of the action seems to be outside of our model. Finally we take a look at a more dynamic model, relating the change in the birthrate to changes in the independent variables. The results are presented in Table 9. We present the estimation results of a set of regression equations explaining the change in the birthrate during 1975-1980 and 1980-1982, respectively, by the change in the independent variables during the same periods. Obviously, this type of model is the most appropriate if one wants to project future changes in fertility due to (predicted) future changes in development and other variables. Table 9: Regression Results; Absolute Change in the Birthrate Due to an Absolute Change in the Independent Variables 1975-1980 and 1980-1982 (standard error in parentheses) 1 22/ MCH Beginning X 1,000 1,000 Rural Staff Rate Heilongjiang Shandong C 1975- 0.036 0.055 0.008 -0.026 0.810 -135.873 12.960 7.136 -20.024 0.506 1980 (0.086) (0.064) (0.159) (0.087) (0.460) (98.669) (2.812) (2.299) 1975- -0.046 0.027 0.092 -0.036 0.269 -0.258 -1.005 4.134 2.698 10.441 0.920 1980 (0.035) (0.026) (0.064) (0.035) (0.190) (41.145) (0.073) (1.326) (1.065) 1980- 0.040 -0.049 -0.034 0.023 0.176 133.335 -4.987 -2.077 4.023 0,346 1982 (0.023) (0.032) (0.035) (0.040) (0.248) (73.503) (1.247) (1.129) 1980- 0.028 -0.028 -0.032 0.007 0.125 126.254 -0.627 -2.002 0.353 10.828 0.521 1982 (0.070) (0.028) (0.030) (0.035) (0.213) (68.083) (0.179) (1.365) (1.190) - 23 - The first equation, based on the R2 corrected for degrees of freedom, seems rather successful: it explains 50 percent of the change in the crude birthrate during 1975-1980. However, the "explanation" of this change is not very useful; the inter-province differences, again, tell the whole story. The birthrate in Heilongliang declined much less than the one in Shandong, which, in turn, declined much less than the one in Sichuan. None of the other variables shows any significant influence. To a sizable extent these inter-provincial differences can be explained by the beginning of the period level of the crude birthrate. When we include this variable it does show a very significant influence (second row, Table 9). The inter-provincial differences reduce in size, but remain significant. Again, none of the other variables shows any significant impact. In some ways, this is a very negative result. The regression shows that, with perfect hindsight, we can explain almost all of the differences in changes in the birthrate (R2 = 0.920). At the same time, we obtain almost no information at all that allows us to project future changes with more confidence. Only two affirmative results emerge: first, the beginning of the period level of the birthrate has a v,ry strong influence, across county birthrate differences are declining. Secondly, fairly large, but unexplained inter-provincial differences exist. The results for- the period 1980-1982 are similar. Inter-provincial differences exist (third row, Table 9), but are much smaller than for the period 1975-1980. Moreover, these differences completely disappear when we include the beginning of the period level of the birthrate (fourth row). This variable turns out to be the only one with a significant effect on changes in the birthrate. The "income" variables, urban/rural differences nor the relative size of the MCH-staff show a significant effect. - 24 - We finally present, in Table 10, the regression results of a model that relates the relative change in the birthrate to relative changes in the independent variables. Since the results in Table 10 tell the same story as those shown in Table 9, we present the results without further discussion. In the previous sections we first presented a simple cross-section model explaining variation in the level of birthrates to variation in the level of independent variables. We then presented models using absolute and relative changes in the birthrate as the dependent variable, and levels, absolute and relative changes in the independent variables as regressors. In the next section we will summarize our results and conclude. Conclusion The main purpose of this paper was to analyze the relationship between fertility and economic development in China. We first showed how easy it is to obtain misleading results from too simple a regression analysis. The results presented in Sections I and II seem to hold a strong warning against the use of cross-section data for making projections, in a highly unstable and heterogeneous environment. In the rest of the paper we focussed directly on the main variable of interest, changes in the crude birthrate. Using per capita agricultural production at the beginning of the period as a development indicator, we find a relatively stable profile for the periods 1975-1980 and 1980-1982 (Figure 1). The profile shows, for 1975-1980, that the largest decrease in the birthrate occurred in low income counties. For 1980-1982, the largest increase occurred in counties with a per capita agricultural production of about 225 yuan. Table 10: Regression Results; Relative Change in the Birthrate Due to an Relative Change in the Independent Variables 1975-1980 and 1980-1982 (standard error in parentheses) X, 2 22 MCK Beginning Heilongjiang Shandong X 2 1,000 1,000 Rural Staff Birthrate D1 D2 C 1975- -0.975 0.102 0.366 -0.030 1.349 -0.017 30.364 28.161 -56.371 0.490 1980 (0.972) (0.216) (0.337) (0.039) (1.048) (0.031) (8.767) ((5.866) 1975- -1.153 0.063 0.392 -0.017 0.584 0.0094 -1.997 14.185 13.966 1.637 0.745 1980 (0.688) (0.153) (0.238) (0.027) (0.752) (0.022) (0.347) (6.806) (4.824) 1980- 2.410 0.715 -0.765 -0.372 0.733 0.344 -45.584 -32.025 33.344 0.386 1982 (1.530) (2.078) (0.557) (0.756) (1.810) (0.211) (11.780) (10.109) 1980- 1.601 0.876 -0.539 -0.410 0.599 0.230 -6.921 -14.808 -7.488 114.200 0.607 1982 (1.238) (1.663) (0.443) (0.605) (1.449) (0.171) (1.589) (11.782) (9.859) - 26 - Figure 1 Absolute Change of Birthrate as a Function of Per Capita Agricultural Output 1975-1980, 1980-1982 ASOLTE CHANGE OF ITMH RATE -S -2- -4- 1975-1980 1980-1982 100 150 200 250 300 350 400 AGRICULTURAL OUTPUT PER CATA - 27 - As we also see in Figure 1, much more important than the marginal impact of the income indicator, is the structural shift that took place between the two periods. In other words, policy measures that directly or indirectly influence the birthrate are much more important in China than the "normal" effects derived from the demand for children. The birthrate level at the beginning of the period turns out to be a very good predictor for the expected magnitude of change in the next few years. The results indicate that across county differences in the crude birthrate are diminishing in China. Though all these results may be of some importance for their own sake, none of them helps us to improve our ability to predict future directions and magnitudes of change in fertility in China. To have repeated cross-section observations on the same units is quite unique in this field. But even in this situation we have to conclude that as long as projections are our main concern, the regression analyses do not teach us much about anything. - 28 - REFERENCES Birdsall, N. and D.T. Jamison. "Income and Other Factors Influencing Fertility in China". Population snd Development Review, Vol. 9, No. 4, 1983. King, T. "Population Policy in China Since 1950 and its Demographic and Economic Implications". Supplemental Paper No. 2, The Health Sector in China, 1984. Statistical Yearbook of China, 1983. Economic Information Agency, Hong Kong, 1983. The Health Sector in China. Population, Health and Nutrition Department, The World Bank, April 1984. Van der Gaag, J. "Commune Health Care in China". Supplemental Paper No 11, The Health Sector in China, 1984. World Development Report, 1984. Oxford University Press, July 1984.