Errata Corrige: “are Muslim Immigrants Different in Terms of Cultural Integration?”

We are thankful to Michael Lundholm and Mahmood Arai for pointing us towards a coding error which invalidates the regressions in our paper. Correcting the code leads to a decrease in sample sizes, though much smaller than Arai et al. (2011) claim based on their “replication”. An appropriate redefinition of the variables and of the model specification allows us to reproduce the substance of the empirical analysis in our original published paper. Although the results are now less clear-cut, our analysis remains essentially unchanged.(JEL: A14, J15) 1. On “Replicating” Bisin et al. (2008) A labelling error in the code invalidates the empirical analysis of our paper on Are Muslim Immigrants Different in terms of Cultural Integration?, published in the Journal of European Economic Association, 6, 445–456, 2008. The error affected our sample selection procedure. We thank Mahmood Arai, Jonas Karlsson, and Michael Lundholm for alerting us to a problem with sample sizes which in turn pointed us to the coding error. In private email correspondence we have provided Arai, Karlsson, and Lundholm with the data and code of a revision of our analysis that we have produced after correcting for the error. In spite of that, the authors decided to submit a note to JEEA which is published in this issue of the journal under the title, “On Fragile Grounds: A Replication of . . .”, Arai et al. (2011). We would like first to comment on the reading of our Introduction by Arai et al. (2011). They find disagreeable our interpretation of the effects of terrorist attacks and riots in Europe on popular opinions regarding Muslim immigration. They are free The editor in charge of this paper was Roberto Perotti. Acknowledgments: We thank Fabrizio Zilibotti for useful comments. Patacchini is a permanent visitor at EIEF and a Research Affiliate at CEPR. Verdier is a Research Fellow at CEPR. Zenou is a Research Fellow at IFN and CEPR. E-mail addresses: [email protected] (Bisin); [email protected] (Patacchini); [email protected] (Verdier); [email protected] (Zenou) Journal of the European Economic Association October 2011 9(5):1012–1019 c © 2011 by the European Economic Association DOI: 10.1111/j.1542-4774.2011.01044.x Bisin, Patacchini, Verdier and Zenou Muslim Immigrants 1013 to believe that terrorist attacks and riots have no effect on popular opinion. Correct logical reasoning does not allow them, however, to conclude that we share the popular opinions we report on. Most importantly, we disagree on the description of their note as a “replication” , as the error invalidates the sample selection at the basis of the empirical analysis of our previous paper. The authors seem to disregard the substance of what the data reveal about the question at hand, that is, the pattern of Muslim integration in the United Kingdom. We explain the issue in the following. The available sample in the Fourth National Survey of Ethnic Minorities 1993–1994 (FNSEM) is dramatically smaller than our coding error had led us to believe in the previously published paper (Bisin et al., 2008). The number of Muslim and non-Muslim after the correction is reduced, respectively, from 2,369 to 2,019 and from 3,594 to 2,639. Arai et al. (2011) report a drop in sample from 5,963 to 1,901 individuals. This is obviously the result of their misguided attempt at minimizing the sample, as we document in turn: (i) They code as missing both the genuine non-available and all the unknown categories, whereas, for example, in some crucial dummy variable definitions (i.e., for the dependent variables) it seems appropriate to include the “Can’t say” category in the reference category (defined as 0). For instance, in the definition of “Attitude towards inter-marriage” those answering “Can’t say” to the question “Would you personally mind if a close relative were to marry a white person?” are clearly individuals with no strong ethnic identity and therefore appropriately included in the reference category. (ii) They eliminate all single (not-married) individuals because one control (dummy) variable, Arranged marriage, is constructed on a question answered by married people only. Precisely, the question is the following one: “What part did your parents play in choosing your husband/wife?” and is coded as a dummy taking 1 if the answer is “The parents made the decision”. It is natural to consider instead singles as individuals whose ethnic identity has not been influenced by an authoritative parental decision regarding marriage. Unfortunately, however, sample sizes decrease further in the regression analysis because of missing values in variables. In this respect the “replication” by Arai et al. (2011) is also misguided: they have in fact chosen to reproduce our regressions using the dependent variables and the extensive set of control variables we had employed (when we erroneously worked with a larger sample). For a large fraction of the agents in the sample several values of these variables are simply not available. In addition, in our analysis, we use a different measure of ethnic identity as dependent variable in different regression model specifications. The individuals having missing values on those variables are not the same across variables. Arai et al. (2011) remove observations with missing values on all variables of interest, thus dropping a priori all those individuals. If the objective is an empirical analysis of the data (regarding pattern of Muslim integration in the UK), this is unjustified. Regression results for different dimensions of ethnic identity can quite naturally rely on a different number of observations. 1014 Journal of the European Economic Association We report here, therefore, on our revision of the analysis of Bisin et al. (2008). Results based on the correct sample are less clear-cut and some estimated coefficients lose statistical significance. We interpret these results, however, as essentially confirming our previous analysis in its substance. 2. Definition of Variables and Model Specification To reduce the number of missing observations we have opted for the following changes with respect to our previous analysis, in Bisin et al. (2008):1 (1) We eliminate some controls (those for which missing value or multicollineary problems are more severe), namely “No parents”, “Parents’ physical contacts”, “Parents’ telephone calls”, “Parents’ letters”, “English spoken at home (younger)”, “Discrimination of own ethnicity”. (2) We redefine the education and labour market variables as follows. We define Ukhigher, a dummy taking value 1 if the respondent reports having O-level or Alevel education; we also define Ukdegree, a dummy taking value 1 if the respondent reports having any non-school diploma (including trade apprenticeship) or university diploma; and “Foreign education,” a dummy taking value 1 if the respondent reports having any overseas qualification.We define Employed, a dummy taking value 1 if the respondent reports being in paid work. (3) We use “House owner” instead of “Household income” as proxy for income. We define House owner as a dummy taking value 1 if the respondent reports she/he owns or is buying his/her accommodation. (4) We redefine our measure of “Attitude towards intermarriage” as a dummy taking value 1 if the respondent answers “Yes” to the question “Would you personally mind if a close relative were to marry a white person?”. In the published paper, the measure of “Attitude towards intermarriage” we had used was a dummy having value 1 for those respondents who, after declaring they would mind if a close relative were to marry a white person, also stated they would mind very much. (5) We use “Attitude towards religious composition in school” as our third proxy of ethnic identity, instead of “Attitude towards racial composition in school”. Precisely, we use the following question: “If you were choosing a school for a child, would you prefer a school for children of your own religion, or would you prefer a school for children of any religion?” The answer is coded as a dichotomous variable, taking value 1 if the individual would prefer a school for children of only her/his religion, and 0 otherwise. (6) We define “Time spent in the UK” as “Years since arrival” if the respondent is not born in the United Kingdom and as “Age” if the respondent is born in the United Kingdom. We also add a squared term on both variables to increase the model’s fit. 1. The code and final data set are available upon request at http://www.nyu.edu/econ/user/bisina/ dofile_Errata_JEEA.pdf and http://www.nyu.edu/econ/user/bisina/data_Errata_JEEA.dta, respectively. Bisin, Patacchini, Verdier and Zenou Muslim Immigrants 1015 TABLE 1. Description of data. n.obs: Nonn.obs: Muslim 2,019 Muslim 2,638 Variable Explanation of the variable n.obs. Mean St.dev. n.obs. Mean St.dev Importance of religion In the text 1032 0.75 0.43 1338 0.44 0.50 Attitude towards intermarriage In the text 1034 0.47 0.5