How to head count ethnic minorities: validity of census surveys versus other identification strategies

ABSTRACT

Censuses ask individuals to identify their own ethnicity. Minorities, however, may be reluctant to self-identify; and thus, censuses may underreport minority populations, raising concerns about measurement validity. We identify and measure the extent of this concern by matching census data on Romas in Romania against a nationwide survey of 2800 municipality experts (SocioRoMap). While not perfect, we find considerable overlap between the two strategies. In the cases where the two measures do not match, the density of community networks is the driver for likelihood of non-congruence – but demographics factors and socioeconomic conditions account for the level, i.e. the magnitude of difference between the two estimation strategies. Given the systemic discrimination of Romas, these results are cautiously reassuring. As most countries head into a census-collecting year, this paper offers an empirical strategy for assessing the validity of self-identified numbers. If governments are concerned about measurement, random samples of expert assessments can help validate. Alternatively, policymakers can focus where miscounts are most likely: urban, ethnically diverse, and poor localities.

KEYWORDS:

• Census
• ethnicity
• expert survey
• Roma
• Romania

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 Unlike standard binary logical variables (see Manski 1990), we cannot observe the treatment here. Moreover, given the noise surrounding attribution bias and social desirability bias, there is no reason to assume the treatments are mutually exclusive. We cannot estimate the treatment effect because that is precisely what we cannot observe.

2 Funding was provided by the Norwegian Financial Mechanism 2009–2014 – as part of the “Fighting Poverty” programme (RO25).

3 The vast majority of these municipalities were congruent with zero self-identified and zero other identified Romas. We include them in our analysis because they are not “irrelevant cases” (Mahoney and Goertz 2004, 654–655). Consider two municipalities with no self-identified Romas. In municipality 1, the community expert considers the poor, dark-skinned family of five as Romas (other identification=5). But in municipality 2, the community expert does not count the poor family as Romas (other identification = 0). To understand why the mismatch happens in municipality 1, we must empirically include municipality 2 and explain why there was no mismatch.

4 We define margin of error as the extent of difference between the proportion (census Roma estimate/total population) and the proportion (expert Roma estimate/total population). A 1% margin of error would indicate two proportions are within 1% of one another, e.g., 6.5% and 7%.

5 $H=1-\sum _{i=1}^N{\pi }_i^2$, where Π_i is the proportion of ethnic group i (i = 1, … , N).

6 $Q=1-\sum _{i=1}^N{\left({\displaystyle \frac{0.5-{\pi }_i}{0.5}}\right)}^2{\pi }_i$, where Π_i is the proportion of ethnic group i (i = 1, … , N).

7 We also consider samples where the census numbers are strictly larger (N = 97) and where the SocioRoMap numbers are strictly larger (N = 1121). The results are substantively no different.

Additional information

Notes on contributors

Zsombor Csata

Zsombor Csata, Associate Professor, Sociology and Social Work, Babeş-Bolyai University; Research Fellow, Centre for Social Sciences, Institute for Minority Studies, Hungarian Academy of Sciences ([email protected])

Roman Hlatky

Roman Hlatky, PhD Candidate, Government, University of Texas at Austin.

Amy H. Liu

Amy H. Liu, Associate Professor, Government, University of Texas at Austin.