A survey of Mexican school children and their families

313 M. Binder Economics of Education Review 18 1999 311–325 Fig. 1. Optimal schooling investments and liquidity con- straints. a: No liquidity constraints; different investors face dif- ferent marginal benefits schedules depending on individual schooling differentials, time preferences and working horizons. b: Investment with binding liquidity constraints reduces attained schooling. Family i exhausts its resources and is constrained to stop school at S i . In family j the liquidity constraint does not bind and schooling continues until S. child in this family falls short of optimal desired school- ing. The presence of liquidity constraints is likely to coincide with a higher marginal cost curve, so that desired schooling will be related to realized school attainment. In poor families, parents face both direct and opportunity costs of schooling. In wealthy families, a child’s opportunity wages are likely to be a tiny fraction of the family budget so that, in some sense, the family does not face them. Thus even when liquidity constraints bind, we would expect that desired schooling would be correlated with schooling eventually attained. The foregoing discussion points to the relevance of desired schooling both as an integral part of schooling demand and also in terms of contributing to schooling outcomes, even in the presence of liquidity constraints. What role, if any, might the community have in determining desired schooling? Community conveys the meaning of something common to all members, some- thing shared. The most basic thing that can be shared is a geographic area: families living next to each other share the same external environment, including the same local establishments and institutions, such as schools and churches. Social relations are likely to arise from this sharing, be they friendships or conflicts. Anthropological studies suggest that extra-family friendships and net- works of reciprocal exchange abound in Mexican neigh- borhoods Lewis, 1959; Lomnitz, 1977. While com- munities under this definition may span neighborhoods, I assume here that sources of community effects, both social and institutional, arise within neighborhoods. The community effects literature suggests four poss- ible mechanisms by which neighborhoods might have independent effects on individual outcomes such as desired schooling, apart from and in addition to family and personal effects. First, there may be different market and institutional resources in different neighborhoods. For example, a rural community might have more opportunities for child labor; an urban community might have better schools. The former would raise the marginal cost of schooling; the latter would lower it. Second, com- munities might form the basis for information about labor market opportunities and schools. There may be less of this information available in neighborhoods where few have studied beyond primary school Wilson, 1987. This mechanism would raise marginal schooling costs by increasing the costs of collecting information. Third, communities may provide social networks that are useful in locating jobs Montgomery, 1991. Marginal benefits might be higher in neighborhoods with good connections. Finally, peer effects, also discussed in the literature as epidemic theory Crane, 1991 and tipping models Schelling, 1978, suggest that people behave as the majority of their peers do, regardless of family back- ground. With peer effects, there may be social costs to pursuing goals that are not the norm, or added benefits from pursuing goals that are. The remarks of the mother quoted at the beginning of this article are emblematic of a neighborhood peer effect.

3. A survey of Mexican school children and their families

From March through October of 1993 I administered a survey of desired schooling, educational attainment, and child and family traits to 341 fifth-grade students and their parents in nine primary schools in Guadalajara and Arandas in the state of Jalisco, and in the border city of Tijuana in Baja California. The populations of these cit- ies share a regional culture, as many residents of Tijuana are from Jalisco and adjoining states. These states com- prise Mexico’s west, and are closely integrated economi- cally and through internal migration de la Pen˜a, 1986. The local economic orientations of Guadalajara, Arandas and Tijuana, however, are distinct. Guadalajara is a 314 M. Binder Economics of Education Review 18 1999 311–325 manufacturing center of domestically distributed con- sumer non-durables with a population of three million. Arandas is a small city of 30 000 that is a commercial center for the agricultural activity that surrounds it. Tijuana is a center of export-oriented maquiladora plants with 750 000 residents 4 . Students in these different cities face distinct labor markets, although the possibility of migration especially from Arandas to Guadalajara may blur the distinction 5 . The fifth grade was chosen as a compromise between two methodological concerns. First, since the purpose of the survey was to explore determinants of schooling among those with the lowest educational attainment, and drop-out rates are high even in primary school, students needed to be relatively young. The primary school efficiency rate—the ratio of graduates to entrants six years earlier—was only 55 in the 1990–91 school year, as reported by the Secretarı´a de Educacio´n Pu´blica, 1986, 1991. Second, students also needed to be mature enough to have some idea of their future plans and the ability to report accurately on basic information about their households. The sample, then, is restricted to chil- dren who have not dropped out of school as of the fifth grade and therefore contains potential sample selection bias against those with less schooling. The 1990 Census reports that 13 of children 15–19 years of age had attained four years of schooling or less. This figure com- bines urban and rural residents, and probably overesti- mates low attainment among urban children. The sample schools were selected by their location in neighborhoods of varying income levels. Since all chil- dren in a randomly selected fifth grade class were inter- viewed 6 , and children typically attend the school closest to them, the sample represents neighborhood families with school-age children. Most parents answered a ques- tionnaire sent home with their children; we directly inter- viewed 94 non-responding parents in their homes. Table 1 compares the characteristics of households who responded independently with those who were visited. Parents in the visited sample were poorer and less edu- cated than those who answered independently. They were also more likely to be migrants. Both parents’ and children’s desired schooling in visited households were more than a year lower than desired schooling in inde- pendently responding households. We were unable to collect information for the parents 4 Populations are drawn from the Jalisco and Baja California state volumes of the 1990 Mexican Census of Population and Housing. 5 Most of the Arandas children had relatives in Guadalajara, which is only a two-hour bus ride away. 6 I was the sole interviewer for the schools in Guadalajara and Arandas, and was ably assisted by Universidad Auto´noma de Baja California student Hector Gutı´errez in Tijuana. of 26 of the interviewed children. The bulk of these 11 are parents from the private school, where non-respon- dents were not visited at their homes at the request of the school’s director. Although the private school sample is probably less prone to selectivity bias with respect to selection against less educated parents, the sample may have systematically excluded wealthier parents. Table 1 shows that children of non-responding parents desire more schooling on average than the rest of the sample. An additional 30 observations are omitted from the analyses of determinants of schooling desires because they are missing one or more variable values. Table 1 shows that this group tends to have higher desired schooling of parents and children, and more income. If many children with high incomes and high schooling desires are excluded, the estimated effect of income may be biased downward. Other characteristics of the excluded group are not very different from those included in the regressions. The data share broad characteristics with the general urban Mexican population 7 . Fig. 2 compares frequency distributions for a 1988 national sample of urban house- holds Inegi, 1992b with the schools sample by weekly income levels. Income for the national sample is reported as multiples of the minimum salary; the 1993 minimum salary was used in the conversion from minimum salaries to pesos 8 . The distribution from my survey approximates the national distribution, especially for families of mod- est means. It does, however, undersample the upper end of the distribution. National education levels for urban adults also suggest that my sample is not unusual: about half of men and women in the same age groups as the surveyed parents had completed six years or less years of schooling in 1992 Inegi, 1993. Table 2 presents the data by sample schools, which are characterized in the first row by income level. The Guadalajara schools show patterns of rising years of desired schooling with rising income levels. The low- income community sample in the small town of Arandas is indistinguishable in its socio-economic characteristics from the low-income samples in Guadalajara, except for a larger proportion of merchants in the occupational dis- tribution not shown. The Tijuana school samples fall between the low- and low–middle- income samples in Guadalajara in child labor-force participation, household spending, and mother’s education in Table 2. In terms of occupational distribution, Tijuana is also very similar to the low- income Guadalajara samples. A x 2 test cannot reject the null hypothesis that the occupational distributions for the 7 According to the 1990 national census, 71 of the Mexican population is urban. 8 The minimum salary in 1993 was N92, or about US30, for a standard 48-hour work week. 315 M. Binder Economics of Education Review 18 1999 311–325 Table 1 Summary statistics for respondents and non-respondents standard deviation in parentheses Non- Included in determinants Respondents respondents regressions? All Not visited Visited Yes No Sample size 315 221 94 26 285 56 Children’s desired schooling 11.5 11.9 10.7 12.6 11.6 11.9 3.7 3.6 3.7 4.2 3.6 4.2 Parent’s desired schooling 12.1 12.6 11.1 NA 12.0 13.1 3.6 3.5 3.7 3.7 3.4 Household characteristics Household weekly spending a 333 379 226 NA 325 590 401 465 117 386 590 Nuclear family 0.69 0.70 0.68 0.73 0.71 0.61 Both parents present 0.84 0.84 0.83 0.81 0.86 0.71 Number of siblings 5.2 5.0 5.8 5.1 5.2 5.2 2.6 2.6 2.7 2.6 2.6 2.7 Own home 0.75 0.71 0.82 NA 0.75 0.69 Own car 0.31 0.37 0.15 NA 0.31 0.34 Parent characteristics Parents’ age 39.4 39.0 40.4 NA 39.3 41.4 7.1 6.6 7.9 7.0 8.6 Parents’ schooling 5.1 5.7 3.8 NA 5.1 5.5 3.7 3.9 2.6 3.6 4.6 Both parents native 0.13 0.17 0.03 NA 0.13 Father’s salary a 382 421 301 NA 385 362 612 727 234 627 352 Mother’s salary a 162 198 111 NA 161 166 178 218 72 183 105 Parents willing to spend on school out of extra 0.91 0.92 0.89 NA 0.91 0.89 income Parents willing to borrow to finance schooling 0.81 0.81 0.82 NA 0.81 0.77 Child characteristics Age 11.4 11.3 11.8 11.7 11.5 11.4 1.5 1.5 1.6 1.8 1.5 1.5 Male 0.53 0.54 0.52 0.46 0.53 0.52 Native 0.67 0.73 0.52 0.69 0.67 0.67 All statistics between 0 and 1 indicate sample proportions for this characteristic. a In nuevos pesos, which exchanged at about three per US at the time of the survey in 1993. low-income Guadalajara, Arandas and Tijuana samples are the same. But in years of desired schooling, the Tiju- ana children break ranks with the low- and low–middle- income Guadalajara children. As indicated in Table 2, the Tijuana children are very aspiring; desired schooling levels match or exceed those of middle-income Guadala- jara children. A x 2 test rejects the null hypothesis that the Tijuana samples have the same desired schooling dis- tribution as the other low-income samples. 4. Can we measure desired schooling?