Forecasting enrollments for immigrant entry-port school districts

For most school districts, magnitudes of student enrollments and the resident school-age population are governed by distinct, broadly foreseeable factors. Conventional approaches to forecasting enrollment account for these factors, integrating data on past and projected births, families' movement to or away from the community, and patterns of progression through the school district's grade structure. In the most straightforward application, forecasting largely entails fine-tuning certain assumptions and assuring consistency in the input data.

More communities now find themselves absorbing heavy influxes of immigrants, as new migration streams branch out geographically across the nation and recent immigrants gravitate increasingly to suburban enclaves (Alba et al. 1999). Immigrant influx-particularly from Mexicois likely to branch out in the future. Amplifying this point, Massey and Espinosa (1997) document the largely self-sustaining processes at work and note the wide diffusion of the governing factors throughout the Mexican population. For example, they cite national surveys (Camp 1993) indicating that about half of adult Mexicans are related to someone living in the United States and that one-third of all Mexicans have come to this country at some point in their life. Phillips and Massey (2000) further quantify these social ties and the potential for continued emigration latent in the Mexican population. They estimate that at least half of all Mexican men coming of age in Mexico today will migrate to the United States at least once by age 40.

For local forecasters, the demography of U.S. immigrant entry-port communities introduces perplexing uncertainties that lie beyond the reach of conventional demographic forecasting approaches. Among these, two general issues stand out: (1) the heightened potential for channelized migration and (2) the prospect of circular migration. The dynamics of immigrant settlement and resettlement establish family chains and well-defined "beaten paths" that may powerfully focus and perpetuate further population movements (see, for example, Goering 1989). The Mexican economy also is influential: Local economic opportunities at origin shape temporary migrants' motivations and strategies, influencing durations of stay in the United States (Lindstrom 1996). Massey and Espinosa (1997) have thoroughly documented the initiation and perpetuation of Mexico-U.S. migration. Especially relevant here is their finding that "[t]he migration of wives and children and the birth of children in the United States strongly raises the odds of taking additional U.S. trips, documented or undocumented, and strongly lowers the odds of returning to Mexico, especially among those with documents" (p. 988).

The behavioral and sociological mechanisms producing such patterns can be generalized, as Massey, Goldring, and Durand (1994) suggest in their five-stage cumulative model of transnational migration. The particular spatial pattern of settlement may be highly idiosyncratic, however, if tied to historical accident or chance destination choices at an early stage in the process.1 Once established, such patterns may generate spatially concentrated migration systems with highly focused sources of in-migration, or highly focused destinations of out-migration, or both (see Plane and Mulligan 1997). The Hmong refugee population, which is concentrated heavily in Fresno, California, illustrates this point. Hmong tend to migrate as a group (Dunnigan et al. 1997: 157). Given an impetus to move, they would likely be drawn only to those few places nationwide (e.g., St. Paul, Minnesota) with existing Hmong communities or where prominent Hmong leaders already have moved.

Chain migration and channelization have been studied extensively by social scientists (see, for example, Jones 1982; Liu, Ong, and Rosenstein et al. 1991). They are understood sufficiently to suggest plausible scenarios about their potential course; these scenarios could integrate the theoretical framework of Massey et al. (1994) with an understanding of the group in question (see, for example, Haines 1997).

As immigrants become established in local communities, they send their children to local schools. In some circumstances, immigrant settlement also may precipitate a circular-and potentially transnational-pattern of student migration. Immigrants' children enrolled in a local public school system may disappear, then reenroll a year (or a season) later, just as Puerto Rican residents circulated through New York City public schools during the 1960s and 1970s (see Rodriguez 1989). Such a pattern, once entrenched, could gain momentum, channeling increasing numbers of immigrant children to a particular school district renowned for meeting their needs.

As immigrant settlement branches out across the nation, local forecasters at the receiving sites must broaden their vision of possibilities and formulate assumptions to account for those possibilities. In this paper I draw on my effort to do so in one such community as part of a long-range enrollment forecast with a 12-year horizon (1997-2008). I undertook this project for the Santa Ana Unified School District (SAUSD), a large public school system in Orange County, situated in a principal region of immigrant and refugee settlement in southern California. In the hybrid approach described here, conventional grade progression forecasting is broadened to allow for several influences distinctive to immigrant entry-port settings in general. Other forecasters may find this approach a useful point of departure in local situations where margins of uncertainty are wide and assumptions are inherently risky.

In the next section I describe the community and its demographic setting, considering the principal factors that shape enrollment trends in large urban school districts and examining additional distinctive features of SAUSD. The latter include an influx of immigrants and refugees, which has concentrated foreign-born newcomers in the area; a circular pattern of student migration between Santa Ana and nearby Mexico; and a population weighted heavily with children as well as adults of childbearing age and housed at a high level of residential density. In the third section, I illustrate the translation of these factors into specific forecasting assumptions, matching those assumptions to future unknowns and compensating for assumption vulnerability. Here the focus extends beyond prediction to early warning about incipient trends that threaten key assumptions and necessitate a change in course. In the final section, I evaluate the strengths and weaknesses of this approach, based on actual enrollment counts now available for 1997 through 1999. This evaluation provides a post-forecast assessment of errors and links them to the assumptions initially adopted.

DEMOGRAPHIC SETTING

Santa Ana Unified, the seventh largest school district in California, serves a student population of about 56,000. These numbers represent the great majority of resident students in Santa Ana city, a central city of the Orange County PMSA, which had an estimated population of 315,000 in 1999. According to census data, Hispanics accounted for 65% of the city's 1990 population; Asians and Pacific Islanders, another 9%. (Corresponding percentages for Orange County overall were 23% and 10%.) SAUSD's student population is 96% minority (predominantly Latino) as of 1999; two-thirds of the students live in households with incomes at or below the federal poverty level. SAUSD grew explosively during the latter 1980s and early 1990s; since then, growth has moderated.

SAUSD is an exemplar immigrant entry-port school district. Its circumstances and future are intertwined with its regional demographic setting (see Rubin-Kurtzman, HamChande, and Van Arsdol 1996) and with its location in Santa Ana, the first major city north of a Mexican immigration checkpoint on Highway I-5. The District is nearly congruent with Santa Ana and is a conspicuous entry port into American society, as shown by these figures:

Fifty-one percent of the city's 1990 population is foreign-born; one-fourth of these foreign-born persons are recent immigrants who entered the United States between 1987 and 1990;

Forty-nine percent of the adult population and 25% of persons under age 18 are not citizens;

Hispanics make up 65% of the population; Vietnamese (along with smaller numbers of Cambodians, Laotians, and Hmong) make up another 6%;

Fifty percent of the city's Spanish-speaking adult population is linguistically isolated, as are 52% of Spanish-speaking children age 5 to 13;

From 1990 to 1994, total refugee influx into the city exceeded 4,400; in 1994 alone (the latest year estimated), the influx of legal immigrants was nearly 2,500.

By the mid-1990s, many families with children attending SAUSD were residents newly legalized in the aftermath of the Immigration Reform and Control Act CIRCA). Under IRCA, some 66,000 former undocumented immigrants applied from the city of Santa Ana to legalize their status. That number represents one of every 46 applicants nationwide in a city inhabited by just 0.001% of the nation's population.2 A modest but increasing number of entering students originate outside the United States (mostly in Mexico); some, as we shall see, engage in a circular pattern of transnational migration.

Regional Demographic Context

SAUSD's resident population has been shaped by the surrounding regional setting (treated extensively in Allen and Turner 1997; Kling, Olin, and Poster 1991; Waldinger and Bozorgmehr 1996). Until recently, that setting was characterized by rapid population growth fueled by an influx of newcomers drawn to an expanding regional economy. Immigrants (especially refugees) settling in California favored certain Orange County communities, including Santa Ana. Scores of Mexican immigrants from the community of Granjenal, for example, settled in Santa Ana, re-creating the community and adding to SAUSD's enrollments ("In Santa Ana" 1997). In addition, nearly one-fifth (4,417) of the 23,068 refugees coming to Orange County between 1990 and 1994 settled in Santa Ana, adding to the city's large existing refugee population (Demographic Research Unit 1996a).

The settlement pattern of these new immigrants resembles earlier patterns: The majority settle in a few counties to join kin and friends. Orange County is the second most frequently intended county destination of residence in California for immigrants, cited by 9% of all legal immigrants who stated an intention to live in California in the mid-1990s (Demographic Research Unit 1999: table 5). For fiscal 1996, the most recent year measured, persons of Vietnamese and Mexican origin account respectively for 27% and 26% of the 17,598 immigrants intending to reside in Orange County; numerous other nationalities account for the rest.

These data underline a perplexing uncertainty about the future of Santa Ana and its neighboring cities. Immigrant communities, once established, become conspicuous destinations for future immigrants with common origins (as exemplified by the movement from Granjenal to Santa Ana). Santa Ana city, however, has little remaining vacant land zoned residential, and the city's housing stock has not increased since 1990. Even so, newcomers eager to settle near kin and friends manage to secure accommodations, doubling up and pushing residential density (persons per household) ever higher (Myers and Lee 1996).3 As of 1990, Santa Ana had the highest incidence of rental overcrowding in Orange County (Myers, Baer, and Choi 1996:75).

The most detailed portrait of this region can be drawn by using 1990 census data, supplemented with postcensal estimates of population, demographic components of change, and residential density (issued annually by California's Demographic Research Unit). These data highlight several key points as of 1997 (when I made this forecast):

Population in Santa Ana city no longer was increasing despite ongoing countywide growth. The city's population increased at an average annual rate of 1.8% from 1990 to 1993, then stabilized at about 307,000 residents with no further gain from 1994 to 1996. By contrast, Orange County's population grew steadily, averaging 1.4% annually over that six-year period according to the then-latest official estimate (Demographic Research Unit 1997a);

The housing stock has not grown since 1990. The estimated number of housing units has hovered around 75,000 (+/-180), reflecting Santa Ana's nearly built-out status-that is, absence of vacant residential land;

Santa Ana's household configuration reflected extended-family living arrangements. Residential density was extraordinarily high and was still rising, indicating an increasingly widespread "doubling up" among families unable to afford housing. The estimated number of persons per household rose from 4.00 (1990) to 4.19 (1997), far above any other city in Orange County.

Factors Influencing School-Age Population

Several factors determine the current and future number of school-age children living in SAUSD, notably the number of births in preceding years, the relative number of women of childbearing age and the cultural influences on childbearing, and the movement of families into and out of Santa Ana.

Orange County (like California overall) is passing through an era of growth in public school enrollment, fueled by a combination of demographic forces that will enlarge the school-age population generally. Public K-12 enrollments countywide were projected to increase 9.3% between fall 1997 and 2001, and 17.6% by 2006 (Demographic Research Unit 1997b). The intensity of increase, of course, likely would vary locally within the county.

Births. Like other heavily Hispanic immigrant entryport cities, Santa Ana is well endowed with both children and future reproductive capacity. In 1990, 30% of its population was under age 18 (compared with 24% in the rest of Orange County); 52% belonged to the reproductive age range, ages 18-44 (compared with 48% in the rest of the county). In addition to structural factors, Santa Ana's ethnic composition entails distinctive cultural influences conducive to above-average family size, which foster growth in the school-age population. The annual number of births, a key determinant of future kindergarten enrollments, was available for prior years at several geographic scales; the most relevant was births to residents of SAUSD (Orange County Health Care Agency 1996). For future births, however, reliable projections were available only at (or above) the county level (Demographic Research Unit 1996b).

The preferred approach would be to account for births within SAUSD separately by race and ethnicity, and to express future births as a function of the corresponding group-specific projections for Orange County. Such disaggregation is ruled out, however, by inconsistencies in definitions of groups. Instead I aggregated historical birth data from the Orange County Health Care Agency (by census tract of residence) up to the school district level and expressed the data as an annual percentage share of all births countywide through 1994. Projected births for Orange County as a whole (from the Demographic Research Unit)4 provided a plausible basis for deriving the number of future births to SAUSD residents, given an assumption about those percentage shares in future years.

Figure 1 shows the available time series on which that assumption was based: an annual part-whole relationship from 1984 to 1994. In 1994, for example, the 7,686 births to SAUSD residents constituted 15.4% of the 49,880 births countywide. This time series displayed an upward trend, peaked in 1990, and reversed thereafter. By 1993, births within the District (based on the tract-level data) were declining both absolutely and as a proportion of all births countywide. Determining the future values to be assumed for that decreasing percentage is a major uncertainty (addressed below).

Student transfers from neighboring districts. Aggregate enrollments also are influenced by transfers of nonresident students into SAUSD from neighboring school districts and by resident students' transfers out. The numbers in both directions have risen modestly in recent years; recent state legislation should make transfers easier for parents in the future. From 1994 to 1996, the number transferring in (mostly children of parents commuting to jobs in Santa Ana) averaged 458 annually; transfers out averaged 1,128 annually. The net effect of these local transfers, then, is a 670-student reduction in SAUSD enrollments in a typical recent year.

Students of foreign origin. SAUSD collects information systematically on each student who enters or returns to the district's schools. The data collection process (governed by federal reporting requirements) is intended to identify incoming students whose native language is not English and/ or who have ever attended school outside the United States. When an entering student applies, the parent or guardian (or other adult) must state where the student attended school immediately before seeking transfer; whether the student has always attended school within the United States (and, if not, whether the student has ever attended school in the United States); the student's native language; and the languages) spoken in the home. Students applying who are not accompanied by an adult are queried directly. The above information is recorded on a routing slip, which accompanies the student and is captured and tabulated by SAUSD's Registration and Testing Center.

The distinctive foreign origins of SAUSD's students are summarized in Table 1. These data record where students were immediately before entering SAUSD. The information distinguishes (1) transfers from another school district within the United States, (2) any previous schooling outside the United States, and (3) previous schooling marked by movement back and forth between Mexico and the United States.

The data reveal several noteworthy points. First, a sizable share of entering students originated outside the United States (36.1% in 1996-1997, up from 26.8% in 1993-1994), and most come from Mexico. Second, a circular pattern to and from Mexico was emerging, both numerically and proportionally. By 1996-1997, 465 entering students (10.3% of all entering students) were "circulators," moving back and forth between Mexico and SAUSD. Their numbers were distributed across all grades, from elementary through high school (data not shown). Although "circulators" constitute relatively few of the new entrants, the pattern itself appears to be firmly established. Conceivably this pattern could boost SAUSD's future enrollments if it evolved in a direction implied by the cumulative transnational migration model that Massey et al. (1994) posit. Under liberalized immigration conditions, recurrent migration between Mexico and the United States might flourish, channeling increasing numbers of immigrant families with children to any school district with a reputation (in Mexico) for meeting their needs.

Summary

The elements of this demographic setting that were apparent in 1997, when the forecast was made, pointed toward two conclusions. First, certain upper limits were in view. The great majority of Santa Ana's resident school-age children already attended SAUSD; there remained relatively few prospective students who might transfer in from nonpublic schools. Furthermore, the city's population, housed at an extraordinarily high level of residential density, appeared to be leveling off after a halt in the growth of housing. Finally, the upward surge in annual births abated after 1992; decline was expected, with noticeably smaller kindergarten classes beginning in fall 1999.

Second, circular transnational student migration between Mexico and SAUSD was firmly established and apparently was rising. That social mechanism potentially could perpetuate-and perhaps increase-SAUSD's involvement with students from nearby Mexico. Perhaps such students would enroll more permanently; perhaps they would induce still others to become circulators. Furthermore, channelized domestic migration was a possibility for Mexican immigrants and southeast Asian refugees presently living elsewhere in the United States. Such families might be drawn to kin who already live in Santa Ana or neighboring immigrant entry-port communities-notably Westminster, the heart of the world's largest Vietnamese emigre community. These considerations posed uncertainties that made predicting the future a challenge, in which past observations had to be balanced with carefully considered judgments about future possibilities.

ENVISIONING THE FUTURE

Demographic forecasting is both science and art: the science of dissecting change into its constituent processes and the art of making informative assumptions about the future course of those processes (see Ascher 1978, 1981). Immigrant entry-port communities make both aspects challenging and add a further element: unavoidably risky assumptions. The processes at work, as we have seen, may extend beyond the ordinary: One must balance what has prevailed (derived from observation) with what could emerge (based on disciplined speculation). The latter, in turn, suggests the need to extend analytic effort beyond prediction to the recognizable dangers inherent in necessarily risky assumptions.

In this section, I illustrate how the logic behind conventional forecasting can be extended to address the possibilities arising in immigrant entry ports. I formulate assumptions about external context, the regional and citywide factors influencing overall population size; processes increasing or reducing enrollments, including births to district residents, transfers to and from neighboring districts, and circular migration patterns that regularly bring students from outside the United States into the schools; and grade progression rates, which index students' progress through SAUSD's grade structure. I use these assumptions to forecast districtwide future enrollment levels by grade. I develop three separate projection series: a "best-estimate" forecast as well as two alternative projections that set plausible upper and lower limits to this forecast, to define margins of potential uncertainty that the District conceivably might need to accommodate.

Finally, where the future is too uncertain to allow prediction, the question "What will happen?" is replaced by "What could happen, and are we prepared if it does?" An approach well suited to this concern is assumption-based planning (ABP), a technique for strengthening plans that address an uncertain future (detailed in Dewar et al. 1993). ABP shifts the focus from prediction to early warning about incipient trends that could threaten a key forecasting assumption and necessitate a change in course. This technique involves five steps: (1) identifying key assumptions, (2) determining their vulnerability out to some specified time horizon, (3) identifying empirical signposts that would signal an impending breakdown of a vulnerable assumption, (4) devising preventive actions to forestall potential vulnerabilities, and (5) devising hedging actions to compensate for actual vulnerabilities.

External Context

Ongoing regional demographic changes can be expected to transform the school-age population of SAUSD. Official projections of public school enrollments (Demographic Research Unit 1997b) capture the effects of these broad changes at the county level; ordinarily such projections provide a basis for predicting local changes.

The regional economy generates the employment growth that draws many newcomers to Orange County. Santa Ana's population growth appeared to be nearing certain upper limits, as suggested by the extraordinarily high level of residential density and incidence of rental overcrowding, the absence of any growth in housing stock since 1990, and the subsequent cessation of population growth after 1993. These considerations formed the basis for assuming that population growth would cease in Santa Ana (although not elsewhere in Orange County). Nevertheless, future channelized migration to Santa Ana remained a possibility of unpredictable magnitude, as illustrated by the uninterrupted rise in residential density through 1997.

Processes That Change Enrollments

Births. Births are one direct determinant of future enrollment counts. The assumptions formulated here derive from the data in Table 2 showing the number of births countywide and the historical proportions attributable to Santa Ana residents. On the basis of these data, I defined three sets of assumptions:

Forecast: The annual number of births to SAUSD residents from 1997 onward was assumed to equal the number projected for Orange County as a whole multiplied by 0.1604, the SAUSD's average observed share of countywide births over the four most recent years (1991-1994), From 1995 to 1996, I assumed transitional proportions as follows: 0.1566 (1995) and 0.1583 (1996), as listed in Table 2.

Plausible high projection: Here I replaced all "forecast" proportions with 0.1673, the District's peak share of countywide births observed in 1990).

Plausible low projection: Here I replaced all "forecast" proportions with 0.1541, the District's lowest-and also most recent-observed share of countywide births since 1990.

Transfers to or from neighboring districts. Student transfers are another direct determinant of the enrollment count. For SAUSD, negative net transfers in recent years have eased enrollment pressure by about 1.3% (650-700 students). As noted above, an average of 458 students transferred into SAUSD each year, offset by an average of 1,128 transfers out, for an average net change of -670. Will gross transfers continue to balance out in this way in the futureand if so, why? On one hand, severe crowding at SAUSD would limit the possible number of transfers in. Parents seeking to transfer children out, however, would find it easier to do so under newly enacted state legislation. Discussions with SAUSD staff members suggested the following forecasting assumptions (based solely on informed judgment):

Forecast: Assumes that the annual number of transfers in will rise by 60 per year and transfers out will rise by 100 per year, for a net annual change of -40 relative to the prevailing average (-670).

Plausible high projection: Assumes that annual transfers in will rise by 60 per year and transfers out will rise by 60 per year-that is, no net annual change from -670.

Plausible low projection: Assumes that annual transfers in will rise by 60 per year and transfers out will rise by 150 per year-that is, a net annual change of -90 relative to -670.

Circular migration between other countries and SAUSD. Whatever its potential significance, the future course of transnational circular student migration is so highly speculative that no single "forecast" assumption can be justified. Given uncertainties about federal immigration policy and the Mexican economy, the possible expansion of circular movement from Mexico lies beyond the reach of quantitative forecasting. One cannot simply ignore this possibility, however; that would merely amount to a "hidden" forecasting assumption (i.e., a constant, ongoing level of circulation).

As SAUSD's future unfolds, the magnitude of "backand-forth" movement will reveal itself in data that the District compiles each year, which will extend the annual series shown in Table 1. Such data enable us to closely monitor the volume of circulation and to manage this unknowable aspect of the future through assumption-based planning. The magnitude of circular student migration was coupled to two scenarios as an empirical "signpost" (step 3 in ABP) of change in the vulnerability of an important assumption-in effect, as an early warning to prompt a timely change in course (ABP steps 4-5).

Persistence scenario: If the annual number of "back-andforth" entrants remains under 1,000, the underlying migration process is regarded as "persisting." Each year, this numerical barometer could revalidate a key premise (or "planning assumption," in ABP terminology) that circulation will persist without transforming the District into a target destination of circular or channelized family migration.

Channelization scenario: If the recorded number of "back-and-forth" entrants ever exceeds 1,000 in a particular year, and then increases by 10% or more in the following year, the underlying migration process is regarded as "channelizing." The choice of these thresholds (1,000 and 10%) rests on judgment alone; other observers might set them higher or lower. Gauged over two years, this "risky assumption" barometer would provide an early warning that SAUSD might have begun evolving into a target destination of circular or channelized family migration.

So long as the magnitude of circulation revalidated the planning assumption of the "persistence scenario," circular migration would be regarded as manageable. If the magnitude ever invalidated that assumption (the "channelization scenario"), circular migration would be flagged as an impending source of growth capable of disrupting future plans for staffing and facilities. That warning, in turn, would prompt a timely change of course to compensate by (1) revising the forecast in light of the threatened assumption (of no further channelization); (2) hedging actions to compensate for actual vulnerabilities (e.g., rebudgeting, ordering portable classrooms); and (3) preventive actions to forestall further expansion of circulation.

Grade Progression

At the heart of conventional enrollment forecasting methods are assumptions about grade progression rates and the kindergarten (KG) "capture rate"; the latter is defined as the percentage of births in a district in calendar year ft - 5] who enroll in KG in the fall of year t. The first step here is to derive the historical grade progression rates that index the pattern of entry into and progress through a district's grade structure. A detailed understanding of that pattern is essential to formulating sound assumptions about the internal dynamics that will shape future enrollments at each grade.

The student population at one grade level advances to the next higher grade year by year, a process documented in a district's fall enrollment counts for pairs of successive years. Such counts are the basis for deriving the time series of grade progression rates displayed in Table 3, which shows the historical progression pattern at each grade for SAUSD. The logic of grade progression involves one entry grade (kindergarten) for which no preceding "grade" exists, because kindergartners derive from births approximately six years earlier. From this perspective, the number of births six years ago can be regarded as the population of a "pseudo grade" that progresses (with a six-year lag) into KG.

One potential ambiguity is the source of the births. For purposes of estimating, the relevant universe is births occurring to the residents of the school district-in this instance, census tracts within SAUSD's boundaries. The count of such births formed the basis for calculating the birth-to-KG capture rates shown in the top row of Table 3. This rate expresses kindergartners each fall as a percentage of all births to SAUSD residents during the fifth preceding calendar year. For example, the fall 1996 KG count shows that SAUSD "captured" 66.9% of the births to its residents during 1991.

Computing a time series like that shown in Table 3 serves two purposes. First, it displays the patterns at each grade level over time, enabling one to pinpoint changes in a district's internal dynamics. Second, it provides an empirical basis for formulating assumptions about the future. We make the following observations in the case of SAUSD:

First, the birth-to-KG (B-KG) rate declined almost steadily from 1990 to 1996. This suggests that SAUSD's KG enrollment count as a share of births within the District was declining (to 66.9% by fall 1996). Why was this so? Possible explanations are an increasing exodus of young families from Santa Ana and a shift to parochial schools, where child care is more readily available.

Second, the G8-->G9 rates also displayed a nearly steady decline, to 111.9 by fall 1996. Progression rates above 100% typically indicate that the grade in question (here, G9) serves as a gateway into the school system from private or other nondistrict schools. For SAUSD, however, the explanation was more complex. The District places overage students who do not speak English back into grade 9 (a practice that accounts for the elevated G8-->G9 progression rate). The observed decline, over time, in that elevated progression rate might well reflect the District's concerted effort at dropout prevention during the 1990s. The decline, however, was consistent with other possibilities as well: Fewer students transferred into SAUSD at the beginning of high school, or perhaps there were fewer non-English-speaking students at or above the typical age of grade 9 pupils. This latter explanation was supported by the District's internal data showing a significant decline in non-English speakers overall, especially at the G9-G12 level. These insights influenced the forecasting assumptions formulated at this level, as discussed below.

Finally, the G10-->G 11 progression rate increased erratically, from 74.2 in fall 1991 to 85.6 by 1996. Further investigation suggested that this trend probably reflects the District's concerted effort to redesignate overage students.

The above observations illustrate the array of influences that may distort observed grade progression in immigrant entry ports. Extraordinary patterns in grade progression rates signal unknowns that must be investigated and understood, not simply dismissed as statistical irregularities to be averaged out. Accordingly I formulated three grade progression schedules (expressed in Table 4 as percentage rates).

The "forecast" schedule embodied the following assumptions, consistent with standard forecasting practice:

The grade progression rate at each grade level in each future year was assumed to equal the weighted average of rates observed in the four most recent pairs of years (shown in Table 3);

The KG capture rate in each future year was assumed to equal 67.43%, the average capture rate observed in the four most recent years.

The "plausible high" schedule differed from the "forecast" on the following points:

The G10-->4G11 grade progression rate in each future year was assumed to equal 85.60%, the latest observed value of this increasing rate;

The KG capture rate in each future year was assumed to equal 77.77%, the highest capture rate observed in the past five years.

The "plausible low" schedule differed from the "forecast" as follows:

The G8-->G9 grade progression rate in each future year was assumed to equal 111.9%, the latest observed value of this declining rate. The intent here was to account for the decrease in overage nonEnglish speakers placed back in G9;

The KG capture rate in each future year was assumed to equal 63.09%, the lowest capture rate observed in the past five years.

Together these three alternatives define what I judged to be a plausible range of uncertainty, which the District should be prepared to accommodate. Given the margins of uncertainty, the promise of greater precision was unwarranted (and probably self-deceptive).

Summary

Envisioning the future here has entailed analyzing historical patterns of enrollment growth and grade progression, factoring in regional demographic shifts, and addressing several future contingencies that increase the margins of uncertainty about the future. Probing these factors in consultation with knowledgeable District staff members indicated the realistic limits one might place on these uncertainties and the critical thresholds for anticipatory action in the event of accelerating channelization.

POST-FORECAST ASSESSMENT

Revisiting one's assumptions about the future several years into an enrollment forecast can be humbling but also enlightening. A post-forecast assessment helps refine technical judgment, indicates where fine-tuning is warranted, and shows where key information is missing. Thus far, this case has afforded the opportunity to revisit the forecast after three years. In this section I present a diagnostic analysis of sources of error in the forecast, and consider how they illuminate particular assumptions.

Table 5 shows a comparison of actual enrollments with enrollments projected by using the "forecast" assumptions. Two summary measures of error are displayed (see Tayman 1996; Tayman and Swanson 1996, 1999). MAPE (the mean absolute percent error) depicts the average error, regardless of direction, over all grades. MALPE (the mean algebraic percent error) measures any forecast bias-that is, a tendency over all grades to over- or underpredict. Districtwide, the Year 3 forecast (55,346) was 4.34% below the actual (57,857) enrollment in fall 1999. This actual value, however, is within the upper limit (58,024) defined by the "plausible high" projection. The MAPE across grades (4.89%) is influenced heavily by extreme errors at KG, G1, and GIl. The MALPE across grades (-3.87%) documents a tendency to underpredict. For the "plausible high" projection, values of MAPE (2.86%) and MALPE (-0.14%) were superior. The lower error here was due largely to the closer accuracy at KG and G1 (data not shown), which underscores the importance of hedging against a potentially misleading recent downturn in births.

Table 6 evaluates specific assumptions against subsequent actual values. These comparisons help to pinpoint large, systematic departures from reality (as distinct from minor or directionless deviations).

Comparison of Forecasting Assumptions With Outcomes

External context. We posited no further growth in Santa Ana's population, given the absence of growth in housing, the record high level of residential density, and the halt in population growth since 1993. After 1996, however, population growth resumed (although more gradually than in the rest of Orange County). Average residential density also rose steadily, reaching 4.30 persons per household by 1999, as 3% more people than in 1996 crowded into virtually the same number (74,900) of housing units. In retrospect, the mid1990s halt in population growth proved misleading. Despite Santa Ana's extraordinarily high residential density, the hypothesized "upper limit" has yet to be reached in this community of extended-family living arrangements.

Processes changing enrollments. Assumptions here pertain to (1) SAUSD's share of all Orange County births, (2) net interdistrict transfers, and (3) circular migration. The first assumption cannot yet be evaluated and remains a key unknown; data are unavailable. Net interdistrict transfers fluctuated erratically, but the predicted value (-40) thus far matches the two-year average (-39). Circular migration has risen steadily, reaching 515 in Year 1 and 603 in Year 2, but still falls below the ABP "signpost" threshold (1,000); thus it revalidates the original planning assumption.

Grade progression rates. Here, experience to date is revealing and underscores several issues that may characterize immigrant entry-port school districts in general:

First, the assumed birth-to-KG capture rate (0.674) proved to be too low, although its measurement may be confounded by the still-unknown SAUSD share of Orange County births. KG as a proportion of prior births has risen, averaging 0.724, above the forecast level assumed (although still within the "plausible high" limit of 0.778). Problematic here is the likelihood that SAUSD's kindergarten enrollments derive not only from prior births to SAUSD residents but also from births to Mexican-resident families who either settled in Santa Ana or adopted the circular migration pattern from acquaintances.

Second, the assumed G8-->G9 rate thus far has proved too high. The actual average rate has fallen below the "plausible low" limit (1.119) and is trending downward, possibly because of fewer overage non-English-speaking students placed back in G9.

Third, the assumed G9-->G 10 and G10-->G11 rates have proved too low. Actual rates have risen toward 1.0, indicating improved retention; probably this is attributable to the District's recently initiated anti-dropout effort.

CONCLUSIONS As new migration streams branch out geographically across the nation, applied demographers working locally must adapt conventional forecasting approaches to the special situations of immigrant entry-port communities. As seen here, extended-family living arrangements can propel extraordinarily high levels of residential density still higher, even in a nearly built-out city. Also, the dynamics of immigrant settlement and resettlement may powerfully focus or perpetuate further influx and precipitate transnational circulation of students.

The approach illustrated here adds to a growing literature (e.g., Perry and Voss 1996; Swanson and Tayman 1995; Tayman 1996; Tayman and Swanson 1996; Yokum and Armstrong 1996) on adapting population forecasting techniques to distinctive locales in order to enhance their accuracy and usefulness. This hybrid approach combines the logic of conventional enrollment forecasting methods with another technique, assumption-based planning, which compensates for vulnerability of assumptions by establishing decision thresholds for anticipatory action. This approach offers applied demographers a point of departure in local situations dominated by wide margins of uncertainty and inherently risky assumptions.