The Ethnosurvey Revisited: New Migrations, New Methodologies?

Abstract

This article provides a detailed review of the ethnosurvey, a research methodology that has been widely applied to the study of migration for almost four decades. We focus on the application of ethnosurvey methods in Mexico and Poland, drawing on studies done in the former country since the early 1980s and in the latter since the early 1990s (including several post-2004 examples). The second case is particularly relevant for our analysis as it refers to a number of novel migration forms that have been identified in Central and Eastern Europe in the post-1989 transition period. Drawing these studies, we consider the advantages and disadvantages of the ethnosurvey as a research tool for studying international migration. Its advantages include its multilevel design, its blend of qualitative and quantitative methods, its reliance on retrospective life histories, and its multisite data collection strategy. These features yield a rich database that has enabled researchers to capture circular, irregular, short-term, and sequential movements. Its disadvantages primarily stem from its hybrid sampling strategy, which necessarily places limits on estimation and generalizability, and the technical challenges of parallel sampling in communities of both origin and destination. Here we argue that the ethnosurvey was never proposed and should not be taken as a universal methodology applicable in all circumstances. Rather it represents a specialized tool that when correctly applied under the right conditions can be extremely useful in revealing the social and economic mechanisms that underlie human mobility, thus yielding a fuller understanding of international migration’s complex causes and diverse consequences in both sending and receiving societies,

INTRODUCTION

International migration has become a controversial topic in recent economic and sociological debates and a major concern of policymakers worldwide. Strong migration research is often lacking, however, owing to the methodological challenges inherent in studying human movements that cross national boundaries, span diverse cultures, embrace multiple languages, straddle political regimes, and are recorded in different statistical systems. Migration is also difficult to measure because it entails moves of diverse duration, direction, documentation, and circularity. As a result of these complexities, official data on international migration are often missing, unreliable, and biased, conditions that only contribute to the controversy and polarization apparent in both public and scholarly discussions of immigration. All of these issues are of critical importance in case of Central and Eastern European countries as emigration started to play a role there as early as in the 1990s and the inflow of foreigners is still far less numerous than in the case of ‘traditional immigration countries’ or Southern European countries.

Over the years, researchers have sought to overcome these well-known data limitations, with varying degrees of success. One of the best-known attempts—and perhaps among the more successful—is the ethnosurvey methodology, first proposed by Massey (1987a) as a synergistic approach combining qualitative and quantitative methods to gather data in both sending and receiving societies (see also Massey 1993; Massey and Capoferro 2004). The methodology was first applied to study Mexico-U.S. migration in a pilot study of four Mexican migrant communities during 1982 (see Massey et al. 1987). The pilot provided a strong proof of concept that led to the foundation of the Mexican Project (MMP) in 1987, which since then has annually completed ethnosurveys in communities throughout Mexico (see Durand and Massey 2004, 2019), then also in Latin America, Central and Eastern Europe, China and Africa (see also below).

Over the years the ethnosurvey methodology has come to be very widely applied in migration research, and the purpose of this article is to assess its advantages and disadvantages by comparing its performance in two particularly well-studied settings: Mexico and Poland. These two cases are interesting not only because of their similarities (both experienced the shock of integration into global markets and a transition from net out-migration to net in-migration) but also because of their differences (the Mexican and U.S. labor markets have been interconnected for more than a century whereas Polish labor markets only began to integrate with those of Western Europe in the 1990s). A comparative analysis thus affords a systematic assessment of the ethnosurvey’s utility for studying different forms of mobility under different migratory regimes and changing socio-economic background (Massey, Kalter, and Pren 2008; Kaczmarczyk 2011).

In our review of the ethnosurvey’s application in Mexico and Poland, we ask whether it still constitutes a useful methodology in migration studies today, and what its value added is compared to other methodologies. Based on the empirical evidence, we argue that the enthnosurvey remains a powerful methodological tool even when other data sources potentially exist; but its utility depends on the specific contexts in which it is applied, the research questions under investigation, and the availability and quality of the alternative data sources. We believe, however, the ethnosurvey methodology carries several advantages compared to these other sources owing to its ability to capture the wide variety of forms that human mobility takes in a manner that is well-grounded in theory (Massey 1999).

We begin by introducing the assumptions and components of the ethnosurvey and then move on to review the experiences of researchers who have analyzed ethnosurvey data gathered in Mexico and Poland. After discussing the pros and cons of the methodology based on our assessment of the empirical evidence, we consider the ethnosurvey’s efficacy as a tool for studying migratory processes in the world today.

ORIGINS AND ASSUMPTIONS OF THE ETHNOSURVEY

The ethnosurvey explicitly conceptualizes human migration as a complex social, economic, and cultural process that unfolds across space and time in multiple contexts. According to Massey (1987a: 1498):

migration is a process, not an event. Unlike birth and death, which happen once and are bounded in space and time, migration involves at least two points on each dimension. Except on the margins, the definitions of life and death are self-evident and widely shared. However, the definition of a move relies on ambiguous concepts of settlement, residence, and place that are socially constructed and culturally variant. The situation is further complicated because moves may occur more than once, and may encompass a variety of origins and destinations.

The ethnosurvey was developed to overcome well-known deficiencies in migration data in the early 1980s. It was not offered as a universal method to be applied in all cases, but as a flexible tool that could capture the complexities of migration as they unfolded across time within specific social, economic, and cultural contexts. It was put forth more as a supplement than a replacement to other potential data sources. Although some scholars initially saw it simply as a way of gathering reliable data on the status of Mexican immigrants in the United States, from the start it was conceived as a multilateral effort intended to support analyses of migration from the viewpoints of both sending and receiving societies.

Instead of attempting to gather information from dispersed immigrant populations scattered throughout the receiving nation, the ethnosurvey began with intensive data collection efforts focused on specific sending communities, followed by parallel sampling in destination areas connected by social networks to those same communities. In this way, it represented a distinctively transnational view. Such a transnational methodology is not new, of course, but follows in a long tradition of mixed, multilevel, multidimensional research conducted in both sending and receiving communities (see Thomas and Znaniecki 1918; Gamio 1930, 1933; Taylor 1932; Louis 1982; Creswell 2003; Miles and Huberman 1984). The fundamental idea of the ethnosurvey is to combine ethnographic and survey methods to gather information at multiple in space and time, yielding an approach in which:

qualitative and quantitative procedures complement one another, and that properly used, one’s weaknesses become the other’s strength, yielding a body of data with greater reliability and more internal validity than is possible to achieve using either method alone. Survey methods produce reliable quantitative data for statistical analysis, generalization, and replication… Anthropological studies, in contrast, capture the richness of life but sacrifice quantitative rigour. Massey (1987a: 1504).

The importance of this blend of qualitative ethnographic methods with quantitative surveys is signaled by the term “ethnosurvey.” On the one hand, surveys yield standardized data that offer an objective “statistical picture” that allows for generalization and ensures replicability. On the other hand, ethnographic fieldwork sheds light on sociocultural origins and consequences of migration while providing a basis for in-depth analyses of social structures and cultural meanings as they play out in the lives of real people. A critical feature of the ethnosurvey has always been its synergistic and balanced approach to the use of quantitative and qualitative methodologies.

The ethnosurvey sought to overcome a perceived lack of knowledge about the microsocial processes involved in human migration in the early 1980s. It explicitly sought to gather information on the diverse forms in which human mobility is expressed, whether short- or long-term, temporary or permanent, circular or settled, and documented or undocumented. Migration was conceptualized as a dynamic social process in which behaviors, motivations, statuses, and intentions varied over time and were in turn affected by transformations that occurred in places of origin and destination and the global political economy.

The ultimate intent of the design was to link together the various components of the migration process as they unfolded in specific locations across time. Investigators sought to connect the departure context (household and community characteristics) with the migration context (the social organization of travel and border crossing), the arrival context (living and working conditions at points of destination) and the return context (post-migration circumstances in the household and community). This approach demanded an extensive set of questions capable of generating data for a diverse array of social, economic, and demographic variables measured at different levels of analysis at multiple points in time.

Over the years, application of the ethnosurvey methodology has enabled analysts to undertake nuanced, timely, and detailed analyses of the characteristics and behaviors of undocumented as well as documented migrants and has supported a succession of sophisticated statistical analyses detailing the effects of changing policies based on large, reliable samples (Massey and Espinosa 1997; Massey and Capoferro 2004; Massey and Sana 2005; Gentsch and Massey 2011; Massey and Pren 2012; Massey, Durand, and Pren 2016).

FOUNDATIONS OF THE ETHNOSURVEY METHODOLOGY

At this writing, the Mexican Migration Project (MMP) has been in existence for 32 years and has 5,129 registered data users. Its information presently comes from representative ethnosurveys done in 170 Mexican communities located in 24 Mexican states, along with surveys of out-migrants from those same communities in the United States. The database currently includes data on 176,696 individuals in 28,319 households. Although the semi-structured interview guide used by the MMP to gather information has changed over the years in minor ways to reflect shifting circumstances and policies in Mexico and the United States, the basic structure and organization of the ethnosurvey has remained constant and rests on five basic epistemological foundations.

Multimethod Data Collection

A fundamental tenet of the ethnosurvey design is the gathering of both quantitative and qualitative data using a mixed-methods approach. Ethnographic research precedes survey work in order to gather basic information about the community, its history of migration, and the specifics of its social organization, economic structure, and ethnic composition. During this time, project investigators also conduct a census of the community to compile a list of potential dwellings to serve as a sampling frame. Information collected in the preliminary ethnographic phase is used to inform the later application of the ethnosurvey itself. Within each household, the head, spouse, or some other knowledgeable informant is interviewed following a semi-structured interview guide that is open to spontaneous remarks about the topic under consideration. Data from the interviews are, in turn, used to inform additional ethnographic work in the community once the survey phase has ended. The extended ethnographic presence of fieldworkers in the community generates familiarity and trust to yield high response rates.

Representative Multisite Sampling

The original ethnosurvey adopted a blended sampling design that used different methods in different contexts. Communities in Mexico were not randomly selected but purposively chosen to build demographic, social, economic, geographic, and ethnic diversity into the sample over time. Within each community, respondent households were selected using simple random sampling. The household interviews quickly revealed the principal places of U.S. destination for migrants in each community and interviewing teams were sent to those locations several months later to complete additional ethnosurveys among out-migrant households that had settled abroad. Unlike the community sample, however, the U.S. sample was compiled using respondent-driven sampling methods that begin with names and contact information gathered from interviewees and informants in the Mexican community. Although this hybrid sample cannot be assumed to be representative, systematic comparisons have shown the resulting data to match information drawn from nationally representative surveys quite closely (Massey and Zenteno 2000; Massey and Capoferro 2004).

Multilevel Data Gathering

The structure of the ethnosurvey lends itself to the compilation of information at multiple levels, and from the beginning of the MMP individual- and household-level data have both simultaneously been gathered. As the number of communities in the database expanded over time, community-level data also began to be compiled; and since migrants move between two different nations, national-level files were likewise added in order to chart trends in basic social, economic, and demographic variables in the binational political economy. A multilevel approach to data gathering and analysis is consistent with theories that emphasize the importance of understanding the context within which migratory decisions are made (see Stark and Bloom 1985; Fawcett 1989; Massey 1999; Faist 2000). As a result, the MMP database has enabled the estimation of detailed multilevel, longitudinal models for purposes of policy analysis and hypothesis testing (Massey 1987b; Massey and Espinosa 1997; Massey, Durand, and Pren 2016).

Life History Compilation

Given the conceptualization of migration as a longitudinal process that unfolds over individual life cycles and evolves historically within specific communities, a core feature of the ethnosurvey is the compilation of individual life histories, beginning with household heads in 1987 and spouses being added in 1991. In addition, abbreviated migration histories are gathered for all individuals within each household by asking about their first and last international trips and the lifetime number of migrations. The life histories are then used to create person-year files that record events and traits for each respondent from birth to the survey date. The resulting life history files can be matched to historical data compiled at the community and national levels to enable the estimation of dynamic, multilevel analyses (see Massey and Espinosa 1997; Palloni et al. 2001; Massey, Durand, and Pren 2015).

Parallel Sampling

Since migration necessarily connects communities of origin and destination, it is obvious that a full assessment of any migration should gather data on people and circumstances at both locations. Because departure and return are selective processes, migrants interviewed only at the place of origin or location of destination will yield biased samples of migrants’ characteristics, motivations, and behaviors. Data collected only at places of origin exclude households that have settled permanently abroad, whereas data gathered only at places of destination misses migrants who circulate back and forth. Communities also vary widely in the origins and extent of their participation in international migration as well as the nature of that participation (as sojourners, settlers, sporadic, or recurrent migrants). Without parallel sampling at points of both origin and destination, this heterogeneity remains unobserved, making it difficult to capture the full complexity of international migration as it dynamically evolves over time in different locations.

THE ETHNOSURVEY IN MEXICO

Mexican migration to the United States is unique in many ways. Much of the southwestern United States was once part of Mexico and only entered the U.S. after 1848, when the Treaty of Guadalupe Hidalgo suddenly rendered some 50,000 Mexicans “immigrants” with no one ever moving (Jaffe, Cullen, and Bosell 1980). Since then the two nations have been both connected and divided by a 3,000 kilometer border.

Significant cross-border migration began late 19^th Century when U.S. investors financed the construction of Mexico’s railroads, connecting agricultural producers in Mexico directly to markets in the United States (Cardoso 1980). Mass migration dates to 1907 when Japanese immigration to the U.S. was curtailed and American employers began to recruit Mexicans as replacements (Massey, Durand, and Malone 2002). Mexican immigration rose thereafter and surged in the 1920s, only to be cut short after 1929 by the Great Depression and a mass deportation campaign (Hoffman 1974). Mexican labor migration was revived in 1942 by the Bracero Program, a temporary worker program that greatly expanded during the late 1950s to peak at around 450,000 entries per year (Massey, Durand, and Malone 2002).

The U.S. Congress cancelled the Bracero Program at the end of 1964, however, and in early 1965 passed legislation to impose numerical limits on legal permanent immigration from Mexico and other countries in the Americas. With opportunities for legal entry curtailed, Mexicans continued migrating without authorization (Massey and Pren 2012) and from 1965 to 1985 a stable migration system based on the circulation of undocumented migrants evolved (Massey, Durand, and Malone 2002). During this time, estimates suggest that around 85 percent of unauthorized entries were annually offset by departures (Massey and Singer 1995). This was the context within which the first ethnosurveys were conducted in four Mexican communities and their U.S. destination areas during the winter of 1982 and summer of 1983 (Massey et al. 1987).

Although “illegal migration” had become a divisive political and policy issue by the mid-1970s, few reliable facts and data then existed to inform the bitter public debate. During 1978, however, a doctoral student in anthropology, Joshua Reichert, had just returned to Princeton from doing a year of fieldwork in a rural Mexican village where nearly three quarters of the households depended on U.S. migrant earnings for sustenance (Reichert 1981, 1982). Looking over Reichert’s field notes, Douglas Massey, then a postdoctoral fellow at Princeton, was impressed by the specificity and detail they contained about the serial movements of people back and forth between Mexico and the United States in both documented and undocumented status.

After raising money to have the migration data coded and entered into machine-readable form, the two scholars began a collaborative analysis. Their research quickly revealed that the quantified information provided a remarkably accurate and revealing account of the history, trends, and patterns of Mexico-U.S. migration (see Reichert and Massey 1979, 1980). Reichert ultimately went on to different career, but Massey became convinced that a blend of ethnographic and survey methods had the potential to provide a unique window on forms of migration that were not captured in other statistical systems. Upon being appointed to his first faculty position, he wrote a grant proposal to fund a pilot project to test this vision.

The grant was ultimately funded, and in 1982 Massey began a long collaboration with anthropologist Jorge Durand, initially in concert with two other Mexican anthropologists, to design, implement, and analyze an “ethnosurvey” of four Mexican communities—a traditional agrarian village, a commercial agricultural town, a small industrial city, and a working class neighborhood in Mexico’s second largest metropolitan area—all located in west-central Mexico, the historical heartland for migration to the United States.

The four communities were not selected because they were thought to contain a large number of U.S. migrants but to represent different levels of urbanism and industrial development, though it turned out that migrants were indeed present at all four locations. The share of households containing “active migrants” (those migrating within the past two years) varied from just nine percent in the industrial city to 36 percent in the commercial agricultural town, and the share with inactive migrants (those who had migrated to the U.S. more than two years before the survey date) ranged from 20 percent in the metropolitan neighborhood to 39 percent in the commercial agricultural center (Massey et al. 1987).

The end result of the pilot was a monograph that laid out the design and philosophy of the enthosurvey method, established the context for U.S. migration in each community, traced each community’s particular history of migration, described its current pattern of migration to the U.S., outlined the social organization of migration from each place, documented the effects of U.S. migration on household economies in the four settings, examined the socioeconomic consequences of U.S. migration in each community, and described the process of social integration among both sojourner and settler migrants in the United States (see Massey et al. 1987).

The book and several associated articles (Massey 1985, 1986a, 1987b) documented the feasibility and utility of the ethnosurvey method and served as the basis for a grant proposal that proposed to undertake a longer-term collection of data using ethnosurvey methods. The specific aim as to create a database on Mexico-U.S. migration that could be updated each year and be freely made available to researchers. This grant was also funded and from 1987 onward the Mexican Migration Project has annually conducted ethnosurveys in selected Mexican communities to create a public use database that, as already noted, presently contains information drawn from 170 different communities.

To conserve space, Table 1 does not list all 170 communities but instead summarizes data from those sampled by Mexican state of origin (complete data on the full set of communities is available from the project website at https://mmp.opr.princeton.edu/). To date, the MMP has completed community ethnosurveys in 24 of Mexico’s 32 states. The number of communities is notably large in five states (Guanajuato, Jalisco, Michoacan, San Luis Potosi, and Zacatecas) which together constitute the historical heartland for migration to the United States, with migratory traditions that go back to the early 20^th Century (Durand, Massey, and Charvet 2000; Durand and Massey 2003). However, newer sending states such as Puebla, Veracruz, and Yucatan were added to the dataset as their inhabitants became active participants in migration to the United States.

Table 1.

Basic information on the ethnosurveys completed in Mexico 1987–2018.

State	Total Communities Sampled	Averages
		Mexican Population	Survey Year	Sample Size	Refusal Rate	US Sample Size
Aguascalientes	2	15,500	1998	125	0.012	4
Baja California	4	1,301,000	1998	151	0.061	8
Chihuahua	8	76,300	2001	156	0.238	0
Colima	3	4,700	1997	124	0.039	13
Durango	4	13,900	1999	164	0.006	15
Guanajuato	16	19,700	1995	165	0.056	12
Guerrero	4	199,500	1995	138	0.114	6
Hidalgo	2	18,000	2002	153	0.100	0
Jalisco	28	212,600	2000	176	0.089	7
Mexico	8	6,100	2009	161	0.072	0
Michoacán	10	93,400	1994	172	0.064	13
Morelos	9	21,400	2013	161	0.054	0
Nayarit	2	18,500	1990	200	0.020	20
Nuevo Leon	2	137,000	2001	176	0.012	0
Oaxaca	4	67,000	1996	150	0.043	5
Puebla	13	232,500	2009	165	0.047	7
Queretaro	8	2,300	2014	132	0.050	2
San Luis Potosi	12	68,700	1998	143	0.031	7
Sinaloa	3	4,300	1998	151	0.017	6
Tabasco	4	12,000	2013	163	0.035	0
Tlaxcala	3	6,700	2002	140	0.084	10
Veracruz	8	61,750	2004	136	0.113	3
Yucatán	6	9,700	2012	153	0.039	3
Zacatecas	7	28,400	1996	209	0.110	9

Source: Mexican Migration Project.

As can be seen, the size of the Mexican communities ranges from small rural villages to large metropolitan areas and across states the average year of the survey ranges from 1990 to 2014 with a mean 2001. The average sample size in Mexico ranges from 124 to 209 households with a mean of 157, whereas in the U.S. it ranges from 0 to 20 with a mean of around six households. In five of the states, the database contains no U.S. sample at all, reflecting the cost and difficulty posed by parallel sampling, especially in recent years.

Across all 170 communities, the correlation between survey year and the number of U.S. households sampled is −0.43. To some extent the lack of U.S. samples in recent years reflects the fact that several states lie in new sending regions (such as Hidalgo and Tabasco) that do not have well-established branch communities in the United States; or they are in border states that do not have strong traditions of U.S. migration (Chihuahua and Nuevo Leon) Refusal rates in Mexican sending communities are generally low, however, ranging from 0.006 to 0.238 with a mean of just 0.062.

THE ETHNOSURVEY IN POLAND

Poland historically has been a country of emigration. From the middle of the nineteenth century onwards, international migration played an important—and sometimes critical—role in Polish history. Additionally, various migration flows are clearly connected as it is in the case of migration of Poles to Germany driven by a few versions of the guestworker schemes (initiated already in the late 19^th century) and then ethnic ties (i.e. mobility of so-called Aussiedler or ethnic Germans) (Kaczmarczyk, 2005). Large migration outflows prevailed before the First World War and during the 1920s and 1930s but largely disappeared after 1945 owing to political restrictions imposed by the postwar communist regime. International migration began to increase again in the early 1970s owing to the normalization of relations between Poland and the Federal Republic of Germany and the liberalization of controls on cross-border movements, and it surged in the 1980s when an estimated 2.2 million persons migrated, constituting about six percent of the total population (Kaczmarczyk and Okólski 2002, 2008).

Contrary to commonly expressed fears (Layard et al. 1992), during the initial transition to a market economy in the 1990s, the international mobility of Poles declined, rising only slightly at the end of the decade. Poland’s population census indicated that around 0.8 million Polish citizens were outside the country in 2002, comprising around 1.8 percent of the total population. Nonetheless, Poland at that time was still one of the most important migrant-sending countries in Europe, with significant numbers of migrants in Germany, Belgium, the United Kingdom, and the United States, as well as several southern European countries. Importantly, many of the “new” migrants were undocumented. The novelty of migration from Poland in the transition period related not only to its irregular or quasi-regular character but first of all to its temporariness (that created serious statistical challenges as we will show below).

Poland’s 2004 accession into the European Union was a turning point in Polish migration history. The early post-accession years saw a spectacular increase in the scale and dynamics of international migration by Poles, which in the EU context compared only to the migration propensity of Romanians. In the peak year of 2007, the stock of expat Polish migrants was estimated to be around 2.3 million persons or about 6.6 percent of the total Polish population. Out-migration remained at relatively high levels despite Europe-wide economic crisis of 2008, however, and by 2017 Poland’s Central Statistical Office estimated the stock of Poles abroad to be around 2.5 million persons (Kaczmarczyk 2018).

In contrast to previous outflows, the new Polish migrants tended to be young, male, work-oriented, and relatively well-educated. Most were temporary migrants going to Anglophone countries such as Britain, which emerged as the principal destination in the post-accession period. Polish migrants nonetheless have a significant presence in most EU countries (Kaczmarczyk and Okólski 2008; Kaczmarczyk 2011; White et al. 2018). As in previous decades, Polish immigration continue to be driven by disparities in earnings and employment, with geographic mobility facilitated by the emergence of migrant networks and the relaxation of the accession period’s earlier transitional arrangements.

Owing to its transitory, transnational, and temporary nature, recent migration from Poland and other Central and East European nations has been labeled “liquid” or “fluid” (Engbersen et al. 2010; Grabowska-Lusińska and Okólski 2009). Even though a majority of Polish migrants are permanent residents or long-term workers living in their countries of destination, they continue to be undercounted in official statistical systems, posing a serious challenge for migration scholars (Kaczmarczyk 2011). Owing to changes in legal regulations and the favorable cost-benefit ratio connected to international migration, the practice of temporary and circular migration has become ever more firmly established (Stola 1998; Okólski 2012). Unfortunately, neither the available data nor existing theoretical typologies that existed in the early 1990s permitted a full assessment of this new reality, and the inaccurate and ambiguous character of official data is what motivated the application of ethnosurvey methods in Poland at that time. In short, the main idea was to identify a methodology suitable for the assessment of new forms of migration in Poland (and in Central and Eastern Europe in general).

According to Iglicka, Jaźwińska and Okólski (1996) and Jaźwińska, Łukowski and Okólski (1997), investigators preferred the ethnosurvey as a means of gathering data for a variety of epistemological reasons. First, it allowed for the initiation of fieldwork without any real knowledge about Polish migration, information that did not yet exist. In addition, the ethnosurvey’s flexible design could accommodate not only orthodox forms of migration but also the short-term circular movements that are typical during the early phases of global market integration (including massive petty trading). Moreover, the ethnosurvey was explicitly designed to include undocumented migrants – an important category after introduction of the free visa regime between Poland and Western European countries (Stola 1998). Next, one of the major premises was that migration is deeply embedded socially – application of this methodology was meant to not only study migration but also its context, i.e., socio-economic transition. Finally, the ethnosurvey allowed for the tracking of population movements within systems of migration that were rapidly creating new transnational social spaces (Billsborrow and Zlotnik 1995; Pries 1996; Faist 2000).

Given these advantages, investigators at the UN Economic Commission for Europe decided to use the ethnosurvey as the principal methodology for a study not only of migration from Poland but from Ukraine and Lithuania as well, though it was clear from the beginning that some modifications in design were necessary (see Mullan 1995). The most important adjustment pertained to parallel sampling. Contrary to the Mexican case, migration from Central and Eastern Europe during the early 1990s could hardly be described as binational. Rather than consisting of closely-connected and well-established migrant communities in two specific countries, the Polish migration system was multilateral and there were not yet any established out-migrant communities at the points of destination, rendering fieldwork abroad of little value.

Investigators thus developed a tripartite design that was applied only at places of origin. The first phase involved monographic research—the gathering of information about each sample community from archival, documentary, and statistical sources—in order to provide the foundation for a later monograph on each community and its history of participation in international migration. The second phase entailed the application of a representative ethnosurvey within each community to gather data on the size and scale of migration as well the specific behaviors and strategies of migrants, data that would enable the creation of new migration typologies.

The final phase of research was ethnographic fieldwork, done to compile qualitative data to support a more in-depth analysis of issues and patterns uncovered in the quantitative survey. Respondents were selected for in-depth qualitative interviews based on their responses to the survey as well through knowledge gleaned during participant observation. Although the possibility of gathering data abroad was left open as an option for additional fieldwork, it was rarely exercised.

In Poland, ethnosurveys were launched in 1994 by the Centre of Migration Research at the University of Warsaw and initially included four communities, two rural and two urban. Follow-up ethnosurveys were executed in 1996 in two additional urban communities (including Warsaw). Data collection resumed once again in 2007 and 2008, when the ethnosurvey was applied to eight additional communities containing both urban and rural residents, with a final community being surveyed in 2010.

Since the number of communities to be surveyed in Poland was initially quite small, police registers dating to the 1980s were used to select places whose residents were actively engaged in international migration. Unlike the MMP, the Polish ethnosurvey gathered detailed data from all household members, not just household heads. Although each community was randomly sampled, the resulting sample was taken to be representative only of that area with no broader claim to national representativeness. As in the MMP, however, the questionnaire allowed flexibility in the sequence of questions and modules, and both migrant and nonmigrant households were surveyed.

As shown in Table 1, the ethnosurvey database for Poland presently includes data from representative samples of 15,756 individuals in 5,204 households located in 15 communities and ten regions (Kaczmarczyk et al. 2011). It provides unique information on mobility and migration during Poland’s transition period and has permitted the assessment of migratory forms that often escape official statistics. Importantly, applications of the ethnosurvey in Poland occurred throughout the transition period and into the post-accession years to capture changes in patterns and processes of migration over time.

ETHNOSURVEYS AT WORK: GAINS AND CHALLENGES

In the first section, we described in details the very origins of the ethnosurvey as applied to the Mexican migration to the US that resulted in the Mexican Migration Project. In the late 1990s, the MMP investigators themselves sought to build on their success in Mexico and launched the Latin American Migration Project, which applied ethnosurvey methods in other nations throughout the region. Beginning with Puerto Rico in 1998, the methodology was subsequently applied in the Dominican Republic and Paraguay in1999, Costa Rica, Nicaragua, and Haiti in 2000, Peru in 2001, Guatemala in 2004, El Salvador in 2007, Colombia in 2008, Ecuador in 2012, and most recently in Uruguay during 2018.

When the Soviet Union collapsed in 1989 and Eastern Europe began to integrate into western markets in the early 1990s, international migration soon followed and in 1994 the Economic Commission for Europe launched an initiative to apply ethnosurvey methods to document the incipient patterns and processes of movement in Poland, Lithuania, and Ukraine (Mullan 1995; Frejka, Okólski and Sword 1998, 1999). As part of this effort, the Centre of Migration Research at the University of Warsaw in 1994 undertook ethnosurveys in four Polish communities (Jazwinska and Okólski 1996) and followed these with two more in 1996, eight in 2007 and 2008, and a final one in 2010 (Kaczmarczyk et al. 2011).

The China International Migration Project was founded by a former research assistant on the MMP and began in 2002 with a round of data collection modeled on the ethnosurvey in the Province of Fujian (Liang and Zhang 2004; Liang et al 2008). The MAFE project (Migration between Africa and Europe) was established by a demographer who spent a postdoctoral semester observing the MMP. It adapted the ethnosurvey for multi-site comparative research in three African nations in 2005 (Senegal, Ghana, and the Democratic Republic of the Congo) as well as destinations six European countries (France, Spain, Italy, Belgium, Britain, and the Netherlands). Like the MMP, it compiled cross-sectional and life history data at the individual, family, and national levels (Beauchemin and González-Ferrer 2011). The most recent application of the ethnosurvey method came with the Bangladesh Environment and Migration Study, which in 2014 compiled retrospective employment and migration histories from more than 3,000 persons in nine villages located in that nation’s southwestern region (Donato et al. 2016).

The ethnosurvey is only one of several methodologies potentially available to measure human migration (Billsborrow et al. 1997; Vargas-Silva 2012; Carletto et al. 2012). Official registries that record the arrival and departure of international migrants have been used effectively in a variety of studies (see Bauer and Zimmermann 1999; Rotte and Vogler 2000; Hatton and Williamson 1998). Official surveys, both cross-sectional (such as Eurostat’s Labor force Survey—see Dustmann and Frattini 2011; Kahanec 2012; Fic et al. 2011) and longitudinal (such as the German Socioeconomic Panel – see Constant and Massey 2003) have also been employed to good effect, as have dedicated non-governmental surveys such as the World Bank’s Living Standard Measurement Survey (see Azzari and Carletto 2009; Carletto, Davis, and Stampini 2006) and the Chitwan Valley Family Study in Nepal ( Bohra and Massey 2009). A variety of data derived from ethnographic fieldwork have also been deployed to study migration processes (see Iosifides 2011; Carling 2008, 2012; Vullnetari 2012; Dick 2018).

Nonetheless, given the challenging realities of migration research, we argue that the ethnosurvey approach continues to offer clear advantages to migration researchers. In addition to allowing the initiation of data gathering without much prior knowledge, its multilevel, multimethod, time-sensitive approach is well-suited to capturing the context and complexities of international migration during a period of economic restructuring and market transition. Its flexible design is also able to capture the wide variety of forms by which human mobility is expressed, which range from short- to long-term moves, sojourning to settlement, and documented to undocumented movements. These advantages were clearly demonstrated in studies of both Mexican and Polish migration. In the following part of this section, we will discuss thoroughly both the benefits and challenges of ethnosurveys completed in both countries. Our aim is not to focus on technical aspects that have been discussed widely already (Beauchemin 2014; Massey and Capoferro 2004; Liu et al. 2016; Riosmena 2016; Kaczmarczyk and Salamońska 2018) but rather on more general issues that allow for better understanding of the main challenges for the methodology and its impact on contemporary understanding of migration in both countries.

The case of Mexico

The MMP was developed in part to counter extravagant claims about Mexican migration that had become commonplace in the U.S. media since the 1970s.¹ These allegations became part of a broader campaign of demonization aimed at delegitimizing Mexicans in particular and Latin Americans in general, a campaign became known as the “Latino Threat Narrative” (Chavez 2001, 2008). None of the numbers or allegations in the commissioner’s statement was based on empirical evidence, and over the next several decades U.S. migration researchers dedicated themselves to providing an evidentiary base for understanding the true realities of migration across the Mexico-U.S. border with two goals: estimating the true size and characteristics of the undocumented population (see Wasem 2011) and explicating the character, causes, and consequences of the annual inflow of migrants (Donato and Armenta 2011). Although the MMP had little to contribute in estimating the size and characteristics of the undocumented population, studies based on MMP data played a major role in elucidating the social and economic processes underlying the annual movement of migrants back and forth across the border, both in documented and undocumented status (Durand and Massey 2004, 2019).

A turning point in the history of Mexico-U.S. migration was the passage of the Immigration Reform and Control Act (IRCA) in 1986, which launched what would prove to be a three-decade militarization of the border while also conferring legal status on several million former undocumented migrants and criminalizing the hiring of unauthorized workers (Massey, Durand, and Malone 2002). Data compiled by the MMP were central in establishing the character and contours of Mexico-U.S. migration before and after IRCA. Data from the MMP’s ethnosurveys revealed the nature of the migratory system that emerged after the cancellation of the Bracero Program and the imposition of numerical limits on legal immigration from Mexico.

The system of undocumented migration that emerged during the period 1965–1985 was heavily circular, with entries being substantially offset by return moves to yield net inflows that were much smaller than those commonly reported in the media (Massey and Singer 1995; Massey and Espinosa 1997; Singer and Massey 1998). Undocumented circulation was dynamically supported by migrant networks that steadily expanded to create migration-specific social capital that lower the costs and risks of unauthorized border crossing (Massey 1986b, 1987b; Massey, Goldring, and Durand 1994; Massey and Espinosa 1997; Massey and Zenteno 1999; Phillips and Massey 2000; Palloni et al. 2001). Despite the stigmatizing and limiting effects of undocumented status, prior to 1986, a lack of legal papers carried no wage penalty in U.S. labor markets once a migrant’s social, economic, and demographic characteristics were controlled (Massey 1987c).

Although Mexican migration was predominantly circular, ethnosurvey data nonetheless identified a characteristic settlement process by which migrants gradually built up time in the U.S. across successive trips, acquired social and economic ties in destination areas, acquired U.S.-specific human capital, and ultimately brought in family dependents (Massey 1985, 1986a; Massey and Espinosa 1997). Associated with settlement in the United States was a legalization process in which migrants over time acquired social and economic ties to family members and employers in the United States who could act as sponsors for adjustment to permanent resident status, a process that was greatly affected by period-specific immigration policies such as IRCA’s amnesty program (Massey and Malone 2003; Cheong and Massey 2019).

In sum, while undocumented migration was not without its costs and risks, ethnosurvey data clearly showed that prior to 1986 the drawbacks were modest and Mexico-U.S. migration constituted a relatively stable and benign system of transnational movement directed overwhelmingly to three states: California, Texas, and Illinois. At these points of destination in these states, Mexican migrants evinced low rates of settlement and experienced few penalties for participation in U.S. labor markets (Massey et al. 1987).

After the passage of IRCA in 1986, these circumstances radically changed as increasingly restrictive immigration policies were enacted and ever greater amounts of money, personnel, and materiel were allocated to immigration enforcement (Massey and Pren 2012)². The militarization of the border profoundly disrupted the stability of the Mexico-U.S. migration system by driving up the costs and risks of unauthorized crossing, and the acceleration of deportations made life increasingly difficult for migrants within the United States.

The counterproductive effects of these policies were clearly documented and broadcast to policy makers and the public, but with little effect (see Massey 1998, 2003, 2007, 2013, 2017). In the wake of IRCA’s passage, the formerly benign migration system was transformed into something far more exploitative and dangerous. According to a causal analysis done using instrumental variable techniques U.S. policies transformed the Mexico-U.S. migration system in counterproductive ways (see Massey, Durand, and Pren 2016).

The border was militarized in phases, beginning with the two busiest sectors in El Paso, Texas and San Diego California, which had the effect of diverting the migratory flows around those two urbanized areas and into the more remote terrain of the Sonoran Desert and into Arizona. The diversion of undocumented traffic away from California in particular, which prior to 1986 was by far the busiest crossing sector, permanently changed not only the geography of border crossing (Massey, Durand, and Malone 2002), but also reconfigured the geography of final destinations, with the number of Mexican migrants going to California dropping from 63 percent in 1990 to just 28 percent in 2000 (Massey and Capoferro 2008; Riosmena and Massey 2012).

As a result of the diversion of migrant flows into more hostile and dangerous landscapes, the costs and risks of unauthorized border crossing increased dramatically, with crossing deaths rising from 147 in 1985 to peak at 492 in 2005 before dropping back to an average of 368 deaths per year thereafter, causing at least 8,927 migrants to lose their lives along the border between 1985 and 2018, while cost of unauthorized entry jumped from $854 in 1985 to $7,000 in 2017 in constant 2016 U.S. dollars (Massey 2018).

Paradoxically, however, the huge increase in border enforcement had little effect on the likelihood of apprehension at the border. In response to the rising enforcement effort, migrants invested in more sophisticated and better-prepared crossing guides and diverted their crossing routes away from sectors being reinforced by the Border Patrol (Massey, Durand, and Pren 2014). Between 1970 and 2010 the probability of being apprehended during any crossing attempt varied from 0.20 to 0.40 with no particular trend and the likelihood of gaining entry to the U.S. over a series of attempts remained near 100 percent (Massey, Durand, and Pren 2016).

The militarization thus had little effect in deterring Mexicans from initiating undocumented migration, but owing to the rising costs and risks of border crossing, migrants minimized their costs and risks by staying longer in the United States instead of returning home to Mexico to face those costs and risks again in the future. As a result, rates of return migration fell sharply as the border militarized after 1986 (Massey 2015; Massey, Durand, and Pren 2015). With the likelihood of initiating undocumented migration to the United States remaining unchanged by border militarization but with the probability of return migration falling sharply, the net volume of undocumented in-migration increased and the rate of the undocumented population growth accelerated during 1990s and early 2000s, with the population of undocumented residents climbing from around two million persons in 1988 to twelve million persons in 2008 (Massey, Durand, and Pren 2014, 2016; Massey 2018).

The criminalization of undocumented hiring in 1986, when combined with a rising share of unauthorized workers in the labor force and a sharp increase in internal enforcement efforts, changed the structure of labor markets throughout the United States. Whereas before 1986 undocumented migrants earned the same wages as those with documents, afterward a significant earnings differential opened between documented and undocumented Mexican workers, a gap that grew larger as the share of unauthorized migrants increased over time (Phillips and Massey 1999; Massey and Gentsch 2014; Durand, Massey, and Pren 2016).

The ethnosurvey’s retrospective life histories show that the likelihood of undertaking a first undocumented trip from Mexico began to decline after 2000 owing to the earlier decline in Mexican fertility rates, which went from around seven children per woman circa in the 1960s around 2.1 children per woman in 2018. The decline in fertility toward replacement level produced a rapid aging of the Mexican population. From1972 to 2010 the average age of those at risk of taking a first undocumented trip rose from 23 to 46 (Massey, Durand, and Pren 2016).

Although the volume of undocumented immigration to the U.S. began to decline around 2000 because of population aging, it came to a definitive end with the onset of the Great Recession in late 2007. Thereafter the likelihood of initiating undocumented migration fell to near zero and the net flow turned negative. Since that date the size of the undocumented Mexican population has steadily declined, dropping by around 1.5 million persons from 2007 to 2016 (Passel and Cohn 2018).

In the end, studies from the MMP indicate that from 1986 to 2017 the United States spent $62 billion ($2016 USD) in a vain attempt to curtail an undocumented flow that would have ceased of its own accord after 2000 because of Mexico’s demographic transition. In the process, it transformed what had been a circular flow of workers going to three states into a settled population of families in 50 states, resulting in a deterioration of wages and working conditions in local labor markets where unauthorized migrants were prevalent.

Although these findings became well-known to migration researchers, they were lost on policy makers in Washington, D.C., who continued to pursue harsh policies of border enforcement and internal deportation despite the fact that net positive undocumented migration from Mexico ended in 2007 and despite the fact that prior enforcement efforts had produced more rather than fewer undocumented settlers. In many ways, the greatest failure of the MMP is the inability of its data and findings to penetrate the mindset of policy makers and opinion leaders in the United States.

Aside from this moral failing, the greatest methodological challenge to the efficacy of the ethnosurvey in Mexico has been a steady deterioration in the ability to undertake parallel sampling as a core component of the MMP’s design. Prior to 2000, the average U.S. sample contained 11 households and only 20 percent of the Mexican communities lacked a parallel sample in the United States. Since 2000, however, the average U.S. sample size fell to just 2.6 households per community with 85 percent of all Mexican samples lacking a parallel U.S. sample. This deficit is particularly troublesome given that undocumented migration has fallen to near zero and interest increasingly centers on the 5.4 million undocumented Mexicans who continue to remain north of the border.

The principal reason for the deterioration of parallel sample was the rise in white nationalism and anti-Latino sentiment in the United States, which has made individuals interviewed in Mexico much less willing to provide contact and other information about their family members in the United States; and it has made migrants living in the United States much less willing to accede to an interview about their migratory experiences. The rise of the Latino Threat Narrative in public discourse in the United States is well documented (Chavez 2001, 2008; Lee and Fiske 2002; Santa Ana 2002) and the key role of anti-Latino sentiment in motivating white opposition to immigration is well established (Abrajano and Hajnal 2015). At this point, the spread of populist white nationalist movements throughout Europe and the Americas may constitute the single greatest challenge for the use of parallel sampling as a core element of the ethnosurvey design (López-Alves and Johnson 2018).

The case of Poland

The first attempts to apply ethnosurvey methods in Poland were undertaken specifically to overcome the fact that the official population registry (known by its Polish acronym SERP) ceased to exist in the late 1980s as the communist regime disintegrated. As a result, in the early 1990s reliable data on migration did not exist and few scholars had much expertise on the topic. Under these circumstances, the ethnosurvey proved to be a very efficient research tool, enabling early investigators to address issues such as migrant selectivity, documentation, and how migration was linked to the rapidly changing social and economic context without much prior information. Indeed, the very structure of the ethnosurvey allowed for the identification of migration complexity as one of the most distinctive characteristics of the Polish situation.

Since the mid-1990s, the availability of ethnosurvey data has allowed scholars to emphasize interdependencies between population mobility and the socioeconomic transition from communism to capitalism (Stola 1998; Jazwinska and Okólski 2001; Kaczmarczyk 2005). As a consequence, studies on Polish migration have become an integral part of the broader field of research on recent economic development (see Massey, Kalter, and Pren 2008). In contrast to other approaches, the core idea of the ethnosurvey is to assess the behavior both of individuals and the households within which they reside, consistent with theories positing the family as a key migration decision-making unit (e.g., Stark and Bloom 1985; Hammar et al. 1997). This feature of the ethnosurvey enabled analysts to consider Polish migration using the latest theoretical frameworks and conceptual categories.

The ethnosurvey’s household-based system of enumeration and data collection, which included absent as well as present household members, proved to be very effective in capturing the diversity of mobility patterns, especially when analyzed in combination with information gleaned qualitatively using ethnographic methods. The resulting data quickly revealed that the usual statistical methods developed to describe long-term and settled migration were not suitable for the analysis of population mobility during a time of social and economic transition. Using these data, Okólski (2001, 2012) was able to identify four types of migrants that were characteristic of Poland’s transition period: petty traders, contract workers (legal temporary migrants working abroad), settlers, and a distinctive category he labeled “incomplete migrants” – persons whose mobility was omitted in official statistics.

As shown in Figure 1, which displays the frequency of migrants across these four categories in the six communities surveyed through 1996, the relative number of settled migrants (those most likely are captured in official data) tended to be low in most communities (16 percent or less), with the exception of Łubniany (where the share was nearly 47 percent). The large majority of migrants in most communities were either petty traders (in Namysłów) or incomplete migrants (in Mońki, Perlejewo, Nowy Targ, and Warszawa). Most of these temporary, short-term migrants were undocumented and thus escaped registration in the formal statistical system.

Figure 1.

Typology of migrants identified in ethnosurveys 1994–1996 Source: Jazwinska (2001)

In structural terms, the predominance of incomplete migrants and petty traders reflected the relative underdevelopment of nations in Central and Eastern Europe at the time, especially their “under-urbanization” (Okólski 2012). Migrants then tended to be relatively unskilled and from peripheral regions of the country, with low social status and a weak position in Polish labor markets (Kaczmarczyk 2013a). The introduction of the concept of incomplete migrants to capture these informal, short-term movers is one of the principal contributions of early research on migration from Poland. Moreover, somehow contrary to the U.S. case discussed above, this term started to be commonly used by policy makers and migration practitioners (Okólski 2012).

A comparison of data in the 2010 ethnosurvey of Sedziszow Malopolski (see Table 2) with Polish census data from the same region in 2011 found that the former recorded five times more incomplete migrants than the latter, allowing for deeper analysis of Poland’s migratory response to the market transition. During this time migration was pursued more as a survival strategy rather than one of income-maximization or risk management. Including these incomplete migrants in the tally, the ethnosurvey found that 20 percent of all respondents had at least some prior migratory experience, a figure that was 2.5 times that recorded in the census.

Table 2.

Basic information on the ethnosurveys completed in Poland 1994–2010.

Community	Rural-Urban Status	Survey Year	No. of Persons	No. of Households	Percent with Migrants
Łubniany	Rural	1994	676	198	49.5%
Namysłów	Urban	1994	1,211	330	37.6%
Mońki	Urban	1994	831	207	52.7%
Perlejewo	Rural	1994	682	166	57.2%
Warszawa	Urban	1996	4,546	1,753	14.9%
Nowy Targ	Urban	1996	1,145	342	44.2%
Biłgorajski	Both	2007	837	250	43.6%
Koszaliński	Both	2008	572	252	14.3%
Kozienicki	Both	2008	621	250	22.4%
Lublin	Urban	2008	522	200	34.0%
Moniecki	Both	2007	761	249	37.8%
Słupecki	Both	2007	879	253	41.1%
Starachowicki	Both	2007	780	252	31.0%
Zgorzelecki	Both	2007	705	250	24.4%
Sedziszowski Malopolski	Both	2010	988	252	18.8%

Source: Own elaboration based on Iglicka, Jaźwińska and Okólski (1996); Jaźwińska, Łukowski and Okólski (1997); and Kaczmarczyk et al. 2011.

This difference reflects the very strict sampling procedures and the strong training of research team members, who were explicitly instructed to gather data on all household members, including those who were temporarily abroad (Kaczmarczyk et al. 2011). In this way, the ethnosurvey proved to be well-suited to capturing unorthodox forms of mobility that emerged before and during the EU accession, when short-term circular moves constituted a significant share of all the mobility that occurred. Notwithstanding these positive aspects of the ethnosurvey, the methodology also carries certain weaknesses.

Although the relative merits of the ethnosurvey as a mixed-methods approach have been discussed extensively by Kaczmarczyk and Salamońska (2018), here we focus on two problematic features that are particularly demanding in the Polish context: its degree of representativeness and the challenges involved in parallel sampling.

A fundamental methodological challenge that migration researchers all face is that migration is a rare event, contrary to popular beliefs (Billsborrow et al. 1997). In the case of Poland, the EU-accession led to an enormous increase in the scale of migration; but as a fraction of the total population it never reached more than 5–7 percent. Moreover, despite all the consternation about mass immigration worldwide, the share of people currently living outside their country of birth is only around 3.4 percent.

Theoretically the best source of migration data short of a census is a probability survey whose representativeness and reliability can be precisely determined (McKenzie and Mistiaen 2009). Even in a traditional country of immigration such as the United States, the share of foreign-born residents is only around 15 percent, meaning that a nationwide random sample of 1,000 respondents will yield only around 150 immigrants, on average, thus offering little statistical power. For detailed studies of immigrants, either much larger samples must be drawn or a more efficient sampling strategy must be implemented, but these options quickly drive up the costs and technical challenges.

Although nations that maintain an up-to-date population registry can, in theory, avoid these problems, registry data are often not reliable because individuals and households often do not deregister before moving abroad, especially when they perceive their absence as temporary. At the same time, many arriving new immigrants may not register themselves, either because they are unfamiliar with the requirement to do so or because they are in irregular status and do not wish to reveal themselves to host country authorities.

The quality of any sample inevitably rests on the quality of its sampling frame—the list of individuals, households, or other units from which respondents are randomly selected. As just noted, however, migrants are often omitted from registries and are undercounted and frequently out of date in decennial censuses. In studies of migration, researchers rarely have access to a complete list of migrants from which a sample can be chosen, and the lack of a reliable sampling frame requires them to turn to non-probability sampling methods where the likelihood of a unit’s selection into the sample is unknown. The most common alternative nonrandom strategies are those that employ chain referral methods, in which certain respondents are asked to identify and provide contact information on people known to them who possess the characteristic of interest (e.g. being foreign-born). The simplest chain referral method is snowball sampling (that used by the MMP) where respondents are simply asked to name others they know in the population of interest, people who are then tracked down and interviewed if at all possible³.

Against this background, the original ethnosurvey methodology in Mexico assumed a two-stage process for sampling communities in countries of origin. In the first stage, communities were non-randomly selected to capture the experience of migrants coming from different levels of urbanism in traditional regions of out-migration. In the second stage, investigators conducted a comprehensive census that listed all potential dwellings in each community to create a sampling frame from which households were randomly selected. Although this procedure worked well in Mexico, its application in Poland was more challenging since conducting a census of potential dwellings is costly in time and money and the efficiency of sampling is low in large cities.

With random sampling, representativeness depends on three conditions: (1) the completeness, reliability, and validity of the sample frame, (2) the accuracy of the random sampling procedure, and (3) the survey response rate. In Poland, the first two conditions are relatively easy to satisfy. Although in Poland existing lists potentially available as sampling frames tend to be inaccurate or outdated, these problems were overcome by taking an initial census of the survey area. Thereafter, the accuracy of random sampling was assured by careful administration and supervision of survey interviewers.

The most serious problem in the Polish case was the response rate, as shown in Figure 2. In those communities surveyed during 1994–1996, the response rate ranged from a low of 50 percent in Warszawa to a high of 78 percent in Mońki, with an average of 64 percent. The response rates were even lower among those communities surveyed during 2007–2010, which ranged from 44 percent in Słupca to 66 percent in Słupecki, with an average rate of just 55 percent. Since the propensity to respond to a survey is always selective on both measured and unmeasured characteristics, the representativeness of the Polish data may be easily given these nonresponse rates.

Figure 2.

Response rates for ethnosurveys completed in Poland 1994–2010 (Source: Kaczmarczyk et al. 2011: 101)

The issue of representatives was studied in an experiment done in conjunction with the 2010 ethnosurvey in Sędziszów Małopolski. Here the total sample was divided into two subsamples. In one of the subsamples respondents were selected using simple random sampling, whereas in the other they were selected nonrandomly using a method known as Adaptive Cluster Sampling (see Thompson 1990; Thompson and Seber 1996; Turk and Borkowski 2005). With ACS, selection into the sample resembles a random walk except that movement between units selected into the sample is not random, but conditional on some variable of interest, in this case prior experience as an international migrant.

ACS improves sampling efficiency whenever the respondents to be sampled are clustered, either socially or spatially. In Poland, international migrants tend to cluster spatially, living near one another in the same general neighborhood (Kaczmarczyk et al. 2011). This clustering occurs because propinquity encourages the formation of interpersonal ties that create strong social networks that are geographically concentrated (Fawcett 1989; Faist 2000). Given the spatial clustering of migrants, we can assume that whenever a randomly selected household contains a migrant, others in the vicinity will also contain migrants. In the case of the ACS implemented in Sędziszów Małopolski, whenever a household contained a migrant, then additional households in the neighborhood were added to the sample until no additional migrant households are encountered.⁴

Comparison of the random sample with the ACS sample done in Sędziszów Małopolski revealed that while the random sample response rate was relatively high at 58 percent and the measured characteristics were close to those observed in census data, persons with a low propensity for migration were overrepresented in the survey while those with a high propensity were underrepresented. As a result, the estimated share of migrants was 18.8 percent in the random sample, but as high as 29.3 percent in the ACS sample.

According to Thompson (1990), the efficiency of ACS compared to random sampling depends on the characteristics of the population under study, the design of the study, and the costs of the study. In later work, Thompson (1997) proposed use of the ACS under conditions where: (1) the population is clustered (i.e. within-cluster variance constitutes a high share of total variance); (2) the population under study is rare; (3) the expected ACS sample is not much larger the random sample; (4) the costs of observing units in clusters are lower than seeking them at random; (5) the costs of observing units lacking the desired condition (i.e. nonmigrants) are lower than observing the costs that do (migrants); and (6) the condition ACS sampling is easy to measure and observe.

These conditions were thoroughly tested in the Sędziszów Małopolski experiment. Monographic research revealed a significant tendency for migrant households to cluster (indeed, one of the new sections of the town was built mainly using migrant remittances) even though migration was a rare event in the larger community. The share of the randomly sampled observations in the total ACS sample increased by around 70 percent compared to the base sample. The costs of both methods were similar and migrant status remained unknown for only two percent of the sample. The main difference was the estimated prevalence of migration—around 19 percent in the random sample versus 29 percent in the ACS sample (Kaczmarczyk et al. 2011). This promising outcome suggests that ACS represents an efficient solution for studies of migrant sending regions.

As in Mexico, parallel sampling proved to be the most challenging feature of the ethnosurvey’s application in Poland, though for very different reasons. Parallel sampling in Mexico was facilitated by the high concentration of Mexican migrants at particular destinations. According to the MMP, over 54 percent of migrating household heads went to California, followed by Texas (17 percent) and Illinois (8 percent); and almost a quarter headed to Los Angeles on their first trip abroad. Figure 3 shows the distribution of Mexican migrants within each community who went to that community’s most popular state of destination.

Figure 3.

Distribution of communities in the MMP by the share of migrants going to the most popular U.S. state of destination for each community (Source: Kaczmarczyk et al. 2011)

As can be seen, in nine percent of the communities over 90 percent of all U.S. migrants went to the most popular destination state; in 17 percent of the communities 80–89 percent of migrants went to the most popular state; and in 21 percent 70–79 percent went to the most popular state. Totaling these three categories, we see that migrants in almost half (47 percent) of the Mexican communities went to the most popular state of destination; and adding the share of communities where 50–59 percent and 50–69 percent of migrants went to the most population destination, the share of communities where at least half of all migrants went to the most popular destination was 75 percent. With such high levels of migrant concentration in a common state of destination, the decision to follow migrant networks to particular points of destination for purposes of parallel sampling and additional ethnographic work is perfectly understandable, though Parrado, McQuiston and Flippen (2005) suggest that within Mexican communities U.S. destinations have diversified in more recent years

In contrast to the Mexican case, Polish migrants – particularly those in the post-accession period – are quite widely dispersed across global labor markets and concentration rates are generally much lower, even if we refer to target countries rather than target states or cities. Figure 4 displays the share of migrants going to the most popular country of destination for each of the Polish communities. Out of the 11 communities sampled, only four communities display a concentration rate above 60 percent. Indeed, the average rate is just 48.5 and seven communities have rates below 50 percent.

Figure 4.

Share of migrants from a given community residing in the most important destination for that community (Source: Jazwinska 2001: 109; Napierala 2008: 187; Kaczmarczyk et al. 2011: 120)

In addition, the level of aggregation matters for the foreign data refers to concentration in countries not urban areas. London is the only identifiable point of urban concentration across all of the different countries. In recent Polish ethnosurveys, attempts have been made to identify linkages between specific origin and destination communities, these efforts proved unsuccessful in cities such as Glasgow, Limerick, and Southampton. Indeed, in the case of Limerick and Glasgow investigators found it easier to identify Polish regions that were not sending migrants than the opposite.

Nonetheless, fieldwork has shown that parallel sampling offers serious benefits, even if extremely difficult, time consuming and costly. Specifically, it allows for capturing those migrants who moved away in whole households and are thus missed by ethnosurveys conducted in sending communities, which is probably the most serious bias in the Polish studies. In practical terms, the only efficient solution may be to start the process at the destination and then to identify the sending communities and households left back home, although neither the sending nor receiving side of the migration process should be prioritized in choosing research sites. Social media might also be used as a possible source of information on the location of out-migrants (Kaczmarczyk et al. 2011).

CONCLUSION AND DISCUSSION

The methodology of the ethnosurvey was developed in the early 1980s as a mixed-methods approach to studying Mexican migration to the United States, but since then studies modeled on the ethnosurvey design have been conducted in Poland, Lithuania, Ukraine, the Dominican Republic, Paraguay, Costa Rica, Nicaragua, Haiti, Peru, China, Guatemala, Senegal, Ghana, Congo, El Salvador, Colombia, Ecuador, Bangladesh, and Uruguay. In all cases the idea was similar: the methodology borrowed standardized data gathering and structured sampling from social surveys and combined them with the ethnographic targeting of specific communities for participant observation, and in-depth interviewing, and an inventory of community circumstances.

The application of the ethnosurvey methodology in Mexico proved to be very successful in capturing the complexities of Mexico-U.S. migration over the years. Data from the Mexican Migration Project have provided detailed data on how patterns have processes of Mexico-U.S. migration have shifted over time in response to changing circumstances in both nations and as a result of U.S. policy interventions at critical junctures. In academia, these data have been widely used to test hypotheses, build theories, and provide substantive information about the causes, consequences, and processes of international migration.

Although data from the MMP have been influential within academia, they have been largely ignored by U.S. policy makers whose misplaced attempts at restriction and repression successively transformed a circular flow of documented workers into a circular flow of undocumented workers, then into large and growing population of undocumented settlers, and finally into mixed-status population of U.S. citizen children with undocumented parents combined with a circular flow of documented but exploited workers. Whereas the gap between empirical evidence and policy action represents the principal moral failing of the MMP, its principal methodological failing has been the steady erosion of parallel sampling as a structural feature of the ethnosurvey. Ironically it was the obdurate resistance of policy makers to facts and evidence and their persistence in ever more punitive and repressive policies that produced this outcome. The rise of white nationalist sentiment in both the political and policy arena have undercut the ability of researchers to successfully continue the selection procedures and respondent-driven sampling methods that worked so well to compile samples of settled U.S. migrants in the 1980s.

With 14 communities studied so far, Poland represents the largest and most important application of the ethnosurvey method outside of Mexico, and the resulting data from Poland likewise has been very influential in academia. It has proved very influential in informing academic debates about the causes, consequences, and forms of out-migration from Poland and how they have evolved over the course of that country’s social, economic, and political transformation from an isolated command economy linked to the Soviet Union into a dynamic market economy being part of the European Union and linked to the global market economy.

As in the United States, while ethnosurvey data and facts gathered in Poland have made their mark academically, they have largely been ignored or overlooked by policy makers in EU destination countries such as Britain, Germany or the Netherlands, where “Polish plumbers” came to symbolize the perceived excess of internal EU migrants the figured prominently in the Brexit vote and in the spread of nativist populism across other countries in Western Europe. Moreover, as in the case of the Mexican Migration Project, the Polish data gathering effort faltered especially in the integration of parallel sampling into the ethnosurvey design. During the early stages of Poland’s migratory transition, this issue stemmed from the fact that initially there were few settled communities of Polish migrants in other countries. Later parallel sampling faltered because of the wide dispersal of Polish migrants across countries and the lack of concentration at specific points of destination raised technical problems that made parallel sampling difficult and costly.

In sum, despite the advantages offered by the ethnosurvey, its methodology nonetheless has clear drawbacks and limitations. First, its hybrid sampling design yields data whose representativeness is open to question. Although sampling within communities is random and therefore representative, the selection of host communities and the surveying of settled out-migrants in foreign branch communities are decidedly non-random. The use of random sampling within communities of origin does mean that together they yield a representative sample even of migrants who were present at the time and place when fieldwork was done. This issue is more and more problematic considering other data sources that are capable of providing less biased picture of the scale and basic structural charakteristics of migration (register data, big data).

The collection of retrospective life histories also introduces recall error into the data, which of course increases the farther back in time one goes. Recall error is mitigated, though not eliminated, by the careful chronological compilation of personal histories using major life events as markers. Dates for events and circumstances are checked for internal consistency during the process of data entry and file construction. Although in both Mexico and in Poland, labor histories have generally found to be quite reliable, there are exceptions, including year-to-year reports of earnings that have proven to be unreliable (and thus the MMP questionnaire consequently only collects this information for the first and last U.S. trips as well as for the current or latest job in the origin community).

Finally, parallel sampling presents serious logistical difficulties and practical challenges that affect response rates as well as data quality. Unlike sending communities, where respondents are located within a limited, circumscribed area, migrants at places of destination tend to be more widely scattered even within a single town or city. Migrants also generally work long hours, leaving little free time and making it difficult schedule interviews; and of course, those migrants who are in irregular statuses tend to be apprehensive, fearful, and untrusting of outsiders. As a result, the financial and temporal costs of interviewing are much greater in places of destination than in communities of origin, and the sample sizes are correspondingly smaller, generally no more than around ten percent of the size of the original sample, yielding less reliable data for samples that are already nonrandom. In addition, as expressions of nativist sentiment and anti-immigrant prejudice have risen over the decades, levels of respondent cooperation have dropped, reducing sample sizes even further and in recent years even precluding the collection of data altogether.

This comparative analysis of the ethnosurvey’s performance in two countries makes it clear that it does not constitute a universal methodology appropriate to all settings and applications. It functions best when other reliable and valid sources of data on migration are not readily available or when existing knowledge about the nature and character of a given migratory flow is limited. It not well-suited to making national-level estimates about the size and composition of a migrant population, nor does it work well in capturing the totality of migratory experiences in large towns and cities. In attempting to assess migration from these contexts, the MMP focuses data gathering on one specific neighborhood, usually one that is working class or poor in composition. Finally, it cannot capture those forms of mobility that are weakly embedded in the social context or not strongly connected to sending communities, which quite commonly the case for skilled and educated migrants. Nonetheless, the ethnosurvey remains quite useful for building a complex understanding the micro social and economic processes of migration and for assessing their relationship to specific social, geographic, and temporal contexts at the individual, household, and community levels. And as such it can efficiently complement other methodologies aiming at assessment of the scale and basic characteristics of migration, including those based on increasingly popular big data.