Ethnocentrism versus group-specific stereotyping in immigration opinion: cross-national evidence on the distinctiveness of immigrant groups

ABSTRACT

While widespread resistance to immigration is well established in advanced democracies around the world, the role of group-specific stereotyping in anti-immigration sentiment has received limited attention. We derive a novel measurement model to assess stereotyping in three Anglo-Saxon democracies – the US, Canada, and the UK – of the modal outgroup in each country (Hispanics in the US and South Asians in Canada and the UK) and Middle Easterners/Muslims. We show that considerable variation exists in degree of stereotyping against the two major immigrant groups. In the US case, we additionally document over-time variation in group stereotyping. In a final step, we demonstrate a relationship between group antipathies and immigration policy views, akin to other policy domains in which public support varies by the ethnic characteristics of policy beneficiaries. To our knowledge, this study is the first to map stereotypes of Muslims in the US in a comparative setting and over time after 09/11, and amongst the first to link views on immigration policies to group-based stereotypes.

KEYWORDS:

• Hispanics
• middle easterners
• group-based stereotyping
• immigration policy opinions

Disclosure statement

No potential conflict of interest was reported by the authors.

ORCID

Stuart Soroka http://orcid.org/0000-0001-7524-0859

Notes

1 See http://www.theatlantic.com/politics/archive/2016/09/donald-trump-and-the-politics-of-fear/498116/.

2 As recently as 1965, immigration quotas based on ethnicity and national origin were the principal criteria for legal admission into the US, resulting in a virtually all-white immigrant population. The ending of quotas resulted in a dramatic shift in the nationality and ethnic profile of incoming immigrants; by 2010 Latin America and Asia accounted for nine of the top ten ‘sending’ regions (Camarota 2012).

3 See https://today.yougov.com/news/2015/11/24/americans-want-register-gun-owners-not-muslims/.

4 In the US, we searched The New York Times and The Washington Post, in the UK, we searched The Guardian and The Times, and in Canada, we searched The Globe and Mail and the Toronto Star. Exact search queries were Immigration OR Immigrant AND (Latino OR Hispanic OR Mexico OR Latin America) for Hispanics in the US (Immigration OR Immigrant) AND (South Asia OR India OR Pakistan OR Bangladesh OR Sri Lanka OR Afghanistan) AND NOT (Muslim OR Islam) for South Asians in the UK and Canada, and Immigration OR Immigrant AND (Middle East OR Muslim OR Islam OR Arab OR Algeria OR Bahrain OR Iraq OR Jordan OR Kuwait OR Lebanon OR Oman OR Qatar OR Saudi Arabia OR Syria OR United Arab Emirates OR Palestine) for Middle Easterners. All counts are unique articles.

5 According to the 2011 Census, Canada has 1,615,145 South Asian immigrants, but only 380,620 Middle Eastern immigrants.

6 For more information on sampling design and sample size, see the appendix.

7 Concretely, this means that we code the median category as non-applicable, i.e. 0. To be sure, we rerun all measurement models having coded the median category as applicable, i.e. 1, and find unchanged results.

8 See the appendix for sample details.

9 Note that all models are fitted independently to each population relying on the EM algorithm with fixed quadrature developed by Chalmers (2012).

10 See the appendix for details of the measurement models.

11 We refer to overall stereotype as the group-specific latent concept.

12 See the appendix for details about this model specification.

13 See the appendix for more details and summary statistics.

14 See the appendix, Equation (A1) for model specification. All model results do not change if we use the collapsed items instead to construct stereotype scores.

15 We use ability estimates from a graded response model applied to the standard resentment questions in the US and to slightly altered versions in Canada and the UK capturing beliefs about minorities. See the appendix for more details.

16 The countries include Iraq, Jordan, Syria, Turkey, Eqypt, Pakistan, Bangladesh, Iran, and Indonesia.

17 See the appendix for summary statistics and more detailed explanations for this variable.

18 While the model fit index of our preferred model can be directly compared to the alternative, two-dimensional model we specified, the comparison to the one-dimensional parsimonious model is more complicated due to the different structure of each model, although the BIC does penalise for models with a higher number of parameters. As a robustness check, we conducted a principal component analysis. We find that the first two principal components explain more than 50% of the variation in the data, more than the rest of the principal components combined, whereas the first principal component explains less than a third of the underling variance in the data. Combined with the comparisons of our fit statistics, we take this as sufficient evidence for the appropriateness of our preferred two-dimensional model over the parsimonious one-dimensional model. In addition, we conduct validity checks of our model, showing that our constructs have reasonable predictive and convergent validity, which we report in the Appendix.

19 The overall BIC of the preferred model is estimated at 89,370.79, while the overall BIC of the more parsimonious one-dimensional model is estimated at 90,099.69 and the overall BIC of the two-dimensional model assuming topic-specific stereotyping is estimated at 89,566.50. See also Raftery (1995) for a discussion on substantive interpretation of BIC comparisons. The BIC values here are based on the additive likelihood of fitting three independent models to three populations (the US, UK, Canada).

20 We use Equation (A1) to compute $\widehat{x_{i,j}}$ relying on three distinctive sets of population-level parameters, and round the fitted values to the full integer.

21 For comparison, the more general one-dimensional model is only able to predict 76.2% of all responses correctly.

22 In 2010, $\sum {\alpha }_{\mathrm{Hispanic}}-\sum {\alpha }_{\mathrm{MiddleEastern}}=3.2$, ${\chi }^2=102.44$, p = 0.00; in 2016 $\sum {\alpha }_{\mathrm{Hispanic}}-\sum {\alpha }_{\mathrm{MiddleEastern}}=-0.1$, ${\chi }^2=0.31$, p = 0.58. Higher values indicate more stereotyped responses. Positive values indicate more negative feelings towards the modal immigrant group, negative values indicate more negative feelings towards Middle Easterners.

23 In Canada, $\sum {\alpha }_{\mathrm{South}-\mathrm{Asians}}-\sum {\alpha }_{\mathrm{MiddleEastern}}=-4.85$, ${\chi }^2=150.45$, p=0.00, and in the UK $\sum {\alpha }_{\mathrm{South}-\mathrm{Asians}}-\sum {\alpha }_{\mathrm{MiddleEastern}}=-4.18$, ${\chi }^2=279.10$, p=0.00. Again, higher values indicate more stereotyped responses. Positive values indicate more negative feelings towards the modal group, negative values indicate more negative feelings towards Middle Easterners. Results are also supported by paired t-tests of raw row-means: in the US, ${\mu }_{\mathrm{Hispanic}}-{\mu }_{\mathrm{MiddleEastern}}=0.05$, t = 7.84, p = 0.00; in the UK, ${\mu }_{\mathrm{South}-\mathrm{Asians}}-{\mu }_{\mathrm{MiddleEastern}}=-0.09$, t = −17.83, p = 0.00; in Canada, ${\mu }_{\mathrm{South}-\mathrm{Asians}}-{\mu }_{\mathrm{MiddleEastern}}=-0.12$, t = −13.78, p = 0.00. Higher values indicate more stereotyped responses. We repeat all analyses having coded the median category as a stereotyped response and find unchanged results.

24 $\sum {\alpha }_{\mathrm{US}}-\sum {\alpha }_{\mathrm{UK}}=5.13$, ${\chi }^2=153.40$, p = 0.00 in the US–UK comparison and $\sum {\alpha }_{\mathrm{US}}-\sum {\alpha }_{\mathrm{CA}}=8.42$, ${\chi }^2=200.80$, p = 0.00 in the US–Canada comparison.

25 $\sum {\alpha }_{\mathrm{US}}-\sum {\alpha }_{\mathrm{UK}}=-1.76$, ${\chi }^2=19.09$, p = 0.00.

26 $\sum {\alpha }_{\mathrm{US}}-\sum {\alpha }_{\mathrm{CA}}=0.51$, ${\chi }^2=0.84$, p = 0.36. Reassuringly, most results are again supported by paired t-tests of raw row-means: For the US–UK comparison, ${\mu }_{\mathrm{Hispanic}}-{\mu }_{\mathrm{SouthAsian}}=0.10$, t = 12.57, p=0.00 and ${\mu }_{\mathrm{MiddleEastern}}-{\mu }_{\mathrm{MiddleEastern}}=-0.03$, t = −4.01, p=0.00; for the US–CA comparison, ${\mu }_{\mathrm{Hispanic}}-{\mu }_{\mathrm{SouthAsian}}=0.18$, t = 18.85, p = 0.00 and ${\mu }_{\mathrm{MiddleEastern}}-{\mu }_{\mathrm{MiddleEastern}}=0.01$, t = 1.31, p = 0.19. Higher values indicate more stereotyped responses. We repeat all analyses having coded the median category as a stereotyped response and find unchanged results.

27 Although note that the empirical distribution of factor scores is not exactly standard-normal.

28 All results are reproduced using the binary items as supposed to the fully graded items used here to construct the stereotype scores.

29 Multicollinearity is a concern because our latent trait scores are heavily correlated by design, i.e. $r_{\mathrm{US}}=0.83$, $r_{\mathrm{UK}}=0.93$, $r_{\mathrm{CA}}=0.92$ . However, the variation inflation factors for these variables are 3.45 for modal group stereotypes, and 3.38 for Middle Eastern stereotypes in the US, 7.41 for modal group stereotypes, and 7.39 for Middle Eastern stereotypes in the UK, and 6.61 for modal group stereotypes, and 6.65 for Middle Eastern stereotypes in the US. These fall within the acceptable range (e.g. Kennedy 1992).

30 We acknowledge that our measures of perceived economic threat could be stronger. For example, we do not have data on perceived net drain of newcomers on the economic resources and whether immigrants will lead to a weaker labour market position for natives; see, for example, Malhotra, Margalit, and Mo (2010).

31 US: the average discrimination parameter of traits associated with economic productivity is 1.09 for Hispanic stereotyping and 1.15 for Middle Eastern stereotyping; the average discrimination parameter of traits associated with cultural assimilation is 1.79 for Hispanic stereotyping and 1.56 for Middle Eastern stereotyping. UK: the average discrimination parameter for traits referencing economic productivity is 1.35 and 1.34 for South Asian and Middle Eastern stereotyping; the averages are 2.04 and 1.95, respectively, for traits relating to cultural assimilation. Canada: the average discrimination parameter for economic traits is 0.53 for South Asian stereotyping and 0.73 for Middle Eastern stereotyping; the averages for traits relating to cultural assimilation are 1.55 for South Asian stereotyping and 1.73 for Middle Eastern stereotyping.

32 Note that while the constraint is mathematically not required, it makes identification more robust (see Rivers 2003 for a thorough discussion on identifiability of IRT models).