METHODOLOGICAL APPENDIXLIVES IN THE BALANCE:ASYLUM ADJUDICATION BY THE DEPARTMENT OF HOMELANDSECURITYAndrew I. Schoenholtz, Philip G. Schrag, and Jaya Ramji-NogalesNew York University Press (2014)The Department of Homeland Security’s United StatesCitizenship and Immigration Services (USCIS) provided us withdata drawn from its Refugee Asylum and Parole System (RAPS),a computerized database used to track the processing of asylumclaims through the affirmative asylum process.1 USCIS staff atone of four regional service centers enter new asylum cases intothe system, inputting data that asylum seekers provide on FormI-589, Application for Asylum and for Withholding of Removal.2As discussed further below, Asylum Office personnel have accessto RAPS in order to update and correct the database withinformation provided by asylum applicants at their interviews.3The data sets provided by USCIS contained 552,760 asylumdecisions rendered between October 1, 1996, and June 8, 2009.Some cases involved both an applicant and a spouse or child whowas already in the United States. To avoid duplication of data,USCIS eliminated the separate entries of dependents from thedatabase. But as discussed further below, the existence of one ormore dependents in the United States was noted in the record ofthe principal applicant.4DHS provided us with fifteen variables for each case: asylumofficer identification number, 5 asylum officer region, case1 U.S. CITIZENSHIP & IMMIGRATION SERVS., AFFIRMATIVE ASYLUM PROCEDURES MANUAL 2-3, ylum/Asylum/2007 AAPM.pdf [hereinafter USCIS ASYLUM MANUAL].RAPS also allows users to determine the status of asylum cases in the immigration courtsystem. Id. at 3.2 Id. at 2-3, 7. USCIS service centers are located in California, Nebraska, Vermont, and Texas.USCIS Home Page, ervice%20Centers(last visited Dec. 12, 2013).3 USCIS ASYLUM MANUAL, supra note 1, at 2-3, 7. In rare cases, generally involvingreapplication after denial and filing by an applicant who had previously been included as adependent on another asylum application, in which direct filing with the Asylum Office is permitted, the data provided on Form I-589 are inputted by Asylum Office personnel. Id. at 7.4 E-mails from Sally Armstrong, Mgmt. & Program Analyst, Asylum Div., Office of Refugee,Asylum & Int’l Operations, U.S. Citizenship & Immigration Servs., to Philip G. Schrag (Apr. 30,2009, and May 5, 2009) (on file with authors).5 These numbers were completely anonymized and do not relate in any way to any actualidentification numbers used by asylum officers. DHS randomly assigned these numbers toasylum officers for use only in the data set that our study examines.1

identification number, 6 filing date, date of entry, status at entry,nationality, gender, religion, ethnic group, age at filing,dependents, representation, final disposition code, and finaldecision date.7DHS provided the data to us organized by the fiscal year in whichthe case was filed. 8 The FY 1998 data set was split into twoparts: those who filed before and those who filed after April 16,1998, the effective date of the one-year deadline. The FY 2009data set included only cases that had been decided as of June 8,2009.We began by excluding data that were not relevant to ouranalysis. First, we eliminated the cases of all Mexican nationals.During the years in question, the vast majority of Mexicanasylum applicants were not genuine asylum seekers; they appliedfor asylum in order for DHS to deny them and refer them toimmigration court, where they could seek another form of relief.9We eliminated 62,568 Mexican cases.Second, we removed from the data all cases that had not yet beencompleted as of June 8, 2009 (the end date for our data set) andcases that had been administratively closed (such as cases inwhich applicants did not appear for their interviews orThese numbers were completely anonymized and do not relate in any way to any actual asylumcase identification numbers. DHS randomly assigned these numbers to asylum cases for useonly in the data set that our study examines.7 The original data are available on the website on which this Methodological Appendix isposted: / LivesInTheBalance/.8 The U.S. Senate website defines the federal government’s fiscal year as ―[t]he accountingperiod for the federal government which begins on October 1 and ends on September 30. Thefiscal year is designated by the calendar year in which it ends; for example, fiscal year 2006begins on October 1, 2005 and ends on September 30, 2006.‖ United States Senate, Glossary, term/fiscal year.htm (last visited Nov. 12, 2010).9 The database that we used excluded 62,568 cases filed by Mexican nationals. According to theU.S. Headquarters Asylum Division of DHS, a very large proportion of these claims were filedonly to enable Mexicans who were residing in the United States to receive referrals toimmigration court, where they could abandon their asylum claims and seek some other type ofimmigration benefit to which they believed they were entitled, such as cancellation of removal.See Andrew I. Schoenholtz, Refugee Protection in the United States Post-September 11, 36COLUM. HUM. RTS. L. REV. 323, 338 n. 62 (2005). Cancellation of removal, which can be grantedby an immigration judge but not by an asylum officer, allows a foreign national to remain in theUnited States if the judge finds that the applicant (a) has resided in the United States for atleast ten years before being served with a notice to appear in immigration court, (b) has been ofgood moral character during that period, (c) has not been convicted of certain offenses, and (d)has a spouse, parent, or child who is an American citizen or lawful permanent U.S. resident andwho would suffer an ―exceptional and extremely unusual hardship‖ if the applicant wasremoved. 8 U.S.C. § 1229b(b) (2006). DHS did not claim, nor do we believe, that there are not anumber of valid Mexican asylum claims during the years covered by the database. But werespect the DHS judgment that many of the Mexican claims were for other purposes. In order tokeep the focus of the study on the adjudication of asylum cases, we have excluded the Mexicanclaims from our analysis.62

discontinued their applications). 10 There were 2,462 cases thathad not been decided and 121,959 cases that wereadministratively closed. We excluded only 102,484 undecided andadministratively closed cases because of category overlap; that is,some cases that remained undecided or were administrativelyclosed were also Mexican cases, and for that reason had alreadybeen eliminated from the database.Third, we eliminated cases in the data set that the Asylum Officedid not adjudicate. For example, the database included the casesof individuals who entered as dependents of successful asylumapplicants. 11 Because these entries did not represent newindividual asylum determinations, we eliminated 2,955 suchcases.Finally, we removed from the data cases that had obviously beenmiscoded in such a way that they could not be used to addressquestions about the impact of the filing deadline. Specifically,these were cases in which the dates of entry were later than thedates of application. Asylum seekers cannot apply for asylumfrom abroad, and DHS informed us that these were coding errors.There were 1,875 cases in the database for which the date ofentry was later than the date of application; we excluded only1,273 for this reason as the others were already excluded on othergrounds. As a result of these exclusions, our data will not matchup with DHS published statistics.12In total, we eliminated 169,280 cases from the data, leaving383,480 cases for our analysis in Chapters Three through Five.In Chapters Six through Nine, we analyzed the cases of the329,336 asylum seekers who applied on time or qualified for anexception to the one-year deadline. In Chapter Ten, we studiedthe 31,635 decisions made by the 221 asylum officers whoattended Asylum Officer Basic Training (AOBTC) classes at theFederal Law Enforcement Training Center (FLETC) betweenJuly 2003 and July 2008.13Cases that had been administratively closed with no decision had final decision codesbeginning with ―C.‖11 Asylum seekers can include their spouses and minor children in their applications forprotection. If their applications are successful, asylum is also granted to these dependents. If anasylee’s dependents live overseas, they can then enter the United States with asylee status. Weidentified these cases using the following status at entry codes: AS1, AS2, AS3, and AY. Furtherdescription and explanation of status at entry codes are available on this book’s companionwebsite, supra n. 7.12 See, e.g., DEP’T OF HOMELAND SEC., 2012 YEARBOOK OF IMMIGRATION STATISTICS (2013),available at earbook.shtm.13 DHS provided us with biographical data only for officers trained within this time framebecause these were the only officers in our database for whom they collected this information. ―Before 2003 [the FLETC] questionnaire was not given to AOBTC students. AOBTC moved fromFLETC after the July 2008 class, and the questionnaire was not given to the two subsequent103

Regression Analyses, Chapters Three through FiveFor the data exploring the one-year filing deadline in ChaptersThree through Five, we ran a total of four binary logisticregressions to confirm the statistical significance of the data wereported in descriptive form. Two of these regressions exploredthe dependent variable of timely filing; the other two usedrejection on the deadline as the dependent variable. Theindependent variables were essentially the same in eachregression, but we ran additional regressions because we groupedthe asylum seeker nationality variable into asylum seekergeographic region of origin. We describe these dependent andindependent variables further below. The timeliness regressionsanalyzed the entire database of asylum seekers, but the rejectionregressions analyzed only asylum applicants who filed more thana year after entry.The independent variables common to all four regressions areAsylum Office region, asylum officer cases decided, the fiscal yearin which the applicant filed for asylum, applicant’s mode of entry,human rights conditions in the applicant’s country of origin,applicant’s gender, applicant’s religion, whether the applicanthad dependents, whether the applicant was represented by anattorney or other representative, and applicant’s age at filing.One of the timeliness regressions also included applicant’snationality, and the other replaced this variable with applicant’sgeographic region of origin, as the latter variable was derivedfrom the former. Similarly, one of the rejection regressions alsoincluded applicant’s nationality, and the other replaced thisvariable with applicant’s geographic region of origin, as again,the latter variable was derived from the first. Both of therejection regressions also included the independent variable lapsebetween date of entry and date of filing.These regression analyses generally confirmed the crosstabulation analyses; we have noted in the text of the relevantchapter when they did not. The results of these regressionanalyses are provided in the Regression Appendix.EstimationWe ran an out-of-sample prediction to determine how manyclasses . . .‖. E-mail from Sally Armstrong, Department of Homeland Security, to Philip G.Schrag (July 1, 2009). 196 of those 221 officers were new hires who were trained shortly afterbeginning service, while the other 25 officers had joined DHS before July 2003. Most of thecases in that database were therefore decided by officers who joined the asylum officer corpsbetween 2003 and 2008. A copy of the FLETC questionnaire appears on this book’s companionwebsite, supra n. 7.4

late applicants barred by the deadline would have been grantedasylum in the absence of the deadline. We performed a binarylogistic regression on untimely and excepted cases, using thedependent variable of grant rate and the following independentvariables: asylum officer’s region, whether the applicant haddependents, the applicant’s filing date, the applicant’s geographicregion of origin, the applicant’s gender, the lapse between theapplicant’s date of entry and date of filing, whether the applicantwas inspected at entry, the applicant’s religion, and whether theapplicant was represented. This regression calculatedcoefficients, which were then applied through a regressionequation to cases rejected because of the deadline. This providedus with the probability that each particular case would begranted asylum. We then used the mean of these values todetermine the percentage of all rejected cases that would havereceived asylum had the one-year deadline not been in effect.We could not include applicants with blank dates of entry inthis estimation, because we could not determine the lapsebetween date of entry and date of filing for these asylum seekers.Regression Analyses, Chapters Six through EightFor the analysis of decisions on the merits in Chapters Sixthrough Eight, we ran two logistic regressions, including one withstandard errors clustered by asylum officer, and one hierarchicallogistic regression on the database of all merits cases, exploringthe dependent variable of grant.The regression modelscontained the following independent variables: regional asylumoffice in which the applicant’s case was heard, the number ofcases previously decided by the asylum officer who heard theapplicant’s case, whether the applicant entered lawfully, theapplicant’s geographic region of origin, the state of political andcivil rights in the applicant’s nation of origin, the applicant’sstated religion, the applicant’s gender,