Abstract
Objectives. To characterize the conflict of sex and gender identity variables in the 2014 Behavioral Risk Factor Surveillance System (BRFSS) sample and examine how this may affect the administration of sex-related health behavior items to transgender participants.
Methods. We conducted a secondary analysis of the 2014 BRFSS gender identity, sex, and sex-related health behavior variables. Twenty states administered the gender-identity variables (n = 154 062), and 691 respondents identified as transgender in the survey (0.4%). We examined conflict among sex, gender identity, and gender-related variables, and compared conflicting and nonconflicting groups across 4 sociodemographic characteristics.
Results. Nearly one third of respondents (27.8%; n = 171) who identified as transgender received sex-specific items that conflicted with their natal sex, thereby reducing the already small subsample of valid responses. There were no significant differences between conflicting and nonconflicting groups on the basis of region, age, race/ethnicity, or type of interview.
Conclusions. Public health surveys should ask respondents to self-identify their sex and gender identity. Interviewer assumptions of respondents’ sex may lead to erroneous collection of sex- and gender-based items, inhibit survey administration, and create problems in data quality.
The field of transgender health research is expanding rapidly, with increasing empirical support for health disparities between transgender and cisgender (nontransgender) people, as well as disparities within transgender populations. Current literature conservatively estimates that 0.53% to 0.6% of the US population or 1 million adults identify as transgender.1–3 Transgender populations experience high levels of social stress in the forms of family rejection, harassment and violence, physical and sexual assault, intimate partner violence, and discrimination, which, in turn, are associated with adverse health outcomes including substance use and abuse, poor mental health, sexual-risk–taking behavior, increased risk for HIV infection, and avoidance of preventative health care and needed treatment services.4–8 However, most of the available data on transgender health disparities come from studies with nonprobability-based sampling strategies, and the collection of gender identity measures in federal probability-based samples, which is critical to better understanding transgender health disparities, remains in its infancy.9–11
The word transgender, though used and understood in many different ways culturally, is used here to describe individuals whose gender identity differs from their assigned birth sex—for example, a person who identifies as a woman and was assigned a male sex at birth is transgender. Although best practices for ascertaining transgender status are still evolving, there is general agreement that biological sex is a necessary element of data collection.12–14 In typical circumstances, measuring biological sex for purposes of discerning transgender status includes asking about the sex that individuals were assigned on their original birth certificate. However, most federal surveys, in asking about respondents’ sex, do not include this question framing, and some telephone-based surveys, such as the Centers for Disease Control and Prevention’s (CDC’s) Behavioral Risk Factor Surveillance System (BRFSS), have used vocal timbre of the respondent as a proxy for sex, an item that is included on all BRFSS surveys. That is, phone-based interviewers could assume the respondent’s sex on the basis of the sound of the respondent’s voice.
According to survey documents, up until 2016, sex could be ascertained by the interviewer at several different points depending on whether it was a cellphone or landline call (Appendix A, available as a supplement to the online version of this article at http://www.ajph.org). For landline calls, there are several times during the random selection process when sex is indicated, in addition to the demographic sex item that landline and cellphone respondents receive (Appendix A). It is unclear how or if sex ascertained during the random selection process is applied to the sex item in the demographics section. Because there is no random selection for cellphone calls, those respondents only receive the sex item from the demographics section (Figure 1). In the survey that the CDC makes publicly available, the instruction to the interviewer for the demographic item of respondent’s sex simply reads, “Ask only if necessary” (Figure 1). To our knowledge, no data are publicly available to ascertain how interviewers determined respondents’ sex—either by asking the respondent or by assuming sex by some other way (e.g., based on vocal timbre). As a consequence, researchers using BRFSS data from years before 2016 cannot determine whether an interviewer determined sex by vocal timbre or asked the participant directly. As of 2016, BRFSS interviewers are required to ask all participants “Are you (1) male (2) female?”
FIGURE 1—
Survey Codebook of Behavioral Risk Factor Surveillance System Indicating the Demographic Variable for Sex of the Respondent: United States, 2014
Although the problems with using vocal timbre as an indicator of sex are amply evident, they are perhaps distilled to crystal clarity when it comes to transgender populations. In 2014, the CDC unveiled an optional survey module for the BRFSS to measure gender identity. For states that opted to administer modules like the gender-identity module, they are administered after all required core modules, including sex and other sex-specific survey items. Participants were asked, “Do you identify as transgender?” Interviewers could give an optional definition of the term transgender to participants who did not understand the question, though there are no data to determine how often the definition was given. If respondents said yes to the first item, they were asked: “Do you identify as (1) female-to-male (2) male-to-female or (3) gender-nonconforming?” Both female-to-male and male-to-female gender identity labels clearly dictate the sex from which a person may be transitioning (e.g., the natal sex of a female-to-male person is female, and they are transitioning to male).
Though logic suggests that the sex identified via this item would match the sex designated at the start of the survey, in reality it may not. There are several possible causes for this misclassification: participants may give erroneous responses to the gender-identity module, or the interviewer’s use of respondents’ vocal timbre to indicate the sex of a respondent could result in misclassification. In this report, we sought to characterize sex and gender identity variables in the 2014 BRFSS sample. Specifically, we explored how sex ascertained through the gender-identity module compared with sex designated in core survey demographic data (e.g., a person with female-to-male gender identity being listed as female sex). In addition, we hypothesized that assumption of sex on the basis of vocal timbre would result in transgender individuals receiving inaccurate sex-specific questions (e.g., female-to-male individuals being asked if they have ever had a prostate-specific antigen [PSA] test).
METHODS
Data are from the 20 US states and territories that fielded the Sexual Orientation and Gender Identity optional module in their 2014 BRFSS surveys. The states and territories were Delaware, Hawaii, Idaho, Indiana, Iowa, Kansas, Kentucky, Louisiana, Maryland, Minnesota, Montana, Nevada, New York, Ohio, Pennsylvania, Vermont, Virginia, Wisconsin, Wyoming, and Guam. We focused our analysis on the individuals who responded to the gender-identity item of the optional module (n = 154 062).
At the start of the survey, respondents were designated as female or male either by the interviewer assuming sex through vocal timbre or, if unclear, by the interviewers asking the respondents if they are male or female (Appendix A).
The core survey of the BRFSS includes 3 sets of sex-specific items asked of all participants. Depending on whether respondents were classified as male or female, they received several sex-specific questions. Respondents designated as male and who were aged 40 years or older received 6 questions about prostate cancer screening. From these questions, we selected one that asked: “Have you ever had a PSA test?” The additional 5 questions were follow-up questions about history of prostate screening or recommendations for screening, and therefore were not necessary to include. There were no sex-specific items available for respondents who were designated as male and were aged younger than 40 years.
Respondents designated as female received 7 questions related to breast and cervical cancer screening. From these 7 items, we selected 3 separate questions about whether the respondent ever had a mammogram, ever had a Papanicolaou test, or ever had a hysterectomy. Women who indicated that they were currently pregnant were not administered the hysterectomy question. The other 4 questions were follow-up questions about history of breast and cervical cancer screening, and therefore were not necessary to include. Respondents designated as female and aged 45 years or younger were also asked if, to their knowledge, they were currently pregnant. Thus, we included 5 dichotomous, sex-specific questions, 1 for persons designated as male and 4 for persons designated as female.
Gender identity was assessed with 2 items: “Do you identify as transgender?” If respondents said yes, they were asked: “Do you identify as (1) female-to-male (2) male-to-female or (3) gender-nonconforming?” Interviewers also recorded responses of “don’t know/not sure” and refusals to answer.
Among the sample of persons who indicated being female-to-male or male-to-female, we created a dichotomous variable indicating whether the sex ascertained via the gender identity module conflicted with the sex assigned at the start of the survey. For example, a designation of male sex at the start of the survey does not conflict with a male-to-female gender identity (nonconflicting), whereas a designation of female sex conflicts with a male-to-female gender identity (conflicting).
We conducted 3 methods of analysis. First, we summarized the prevalence of male and female designation by each category of gender identity. Second, we used the χ2 test of independence among the subsample of conflicting and nonconflicting groups of transgender individuals to compare 4 sociodemographic characteristics: region (i.e., Northeast, South, Midwest, and West), age group (i.e., 18–24, 25–34, 35–44, 45–54, 55–64, and ≥ 65 years), race/ethnicity (i.e., White, Black, Hispanic, other, multiracial), and the type of interview (i.e., landline telephone or mobile phone). We selected these demographic characteristics because of their potential implications on how an interviewer may hear the respondent’s voice (i.e., regional differences in dialect,15 voices tend to deepen with age,16 perceived racial/ethnic differences in intonation,17 and quality of the phone connections during the interview). We weighted all of these analyses to account for the complex sampling design of the BRFSS.
Third, we illustrate the prevalence of responses to the sex-specific questions by sex and gender identity. We elected not to weight the percentages in this analysis because the principal reason for presenting these results was to simply illustrate whether there existed suspect findings (e.g., female-to-male individuals who reported having PSA tests), and weighting these findings would do nothing to clarify their meaning. We conducted all analyses with Stata SE version 14 (StataCorp LP, College Station, TX).
RESULTS
Among our sample, 0.2% (n = 363) identified as male-to-female, 0.2% (n = 212) identified as female-to-male, and 0.09% (n = 116) identified as gender-nonconforming. Of the 691 participants who identified as transgender, nearly a third (27.8%; n = 171) were designated a sex that conflicted with their gender identity, and subsequently received sex-specific items that would be anatomically implausible (Table 1). Specifically, 26.2% (n = 119) of male-to-female respondents were designated as female. Among female-to-male respondents 30.5% (n = 52) were designated as male. In addition, 18.4% (n = 116) of transgender respondents identified as gender-nonconforming, and because this gender identity label does not dictate assigned birth sex, it is unclear whether the sex of these participants was misclassified (Table 1).
TABLE 1—
Prevalence of Sex Designation by Gender Identity, 20 States and Territories: Behavioral Risk Factor Surveillance System, United States, 2014
| Gender Identity |
||||||
| Variable | Male-to-Female (n = 363), No. (%) | Female-to-Male (n = 212), No. (%) | Gender-Nonconforming (n = 116), No. (%) | Cisgender (n = 150 765), No. (%) | Don’t Know (n = 1138), No. (%) | Refused (n = 1468), No. (%) |
| Sex | ||||||
| Male | 244 (73.8) | 52 (30.5) | 55 (52.7) | 62 086 (47.9) | 502 (49.8) | 569 (46.1) |
| Female | 119 (26.2) | 160 (69.5) | 61 (47.3) | 88 679 (52.1) | 636 (50.2) | 899 (53.9) |
Note. Frequencies are unweighted; percentages are weighted. The states and territories are DE, HI, ID, IN, IA, KS, KY, LA, MD, MN, MT, NV, NY, OH, PA, VT, VA, WI, WY, and Guam.
There were no significant differences in conflicting sex and gender identity on the basis of region, age, race/ethnicity, or type of interview (Table 2).
TABLE 2—
Sociodemographics by Conflicting and Nonconflicting Sex and Gender Identity Among Transgender Individuals in 20 States and Territories: Behavioral Risk Factor Surveillance System, United States, 2014
| Variable | Nonconflicting Sex and Gender Identity (n = 404), No. (%) | Conflicting Sex and Gender Identity (n = 171), No. (%) | P |
| Region | .35 | ||
| Northeast | 97 (37.4) | 38 (25.8) | |
| South | 67 (16.2) | 35 (25.2) | |
| Midwest | 168 (39.5) | 62 (40.2) | |
| West | 72 (6.8) | 36 (8.8) | |
| Age group, y | .7 | ||
| 18–24 | 24 (11.7) | 12 (19.8) | |
| 25–34 | 29 (7.4) | 10 (8.9) | |
| 35–44 | 62 (23.1) | 22 (17.8) | |
| 45–54 | 78 (20.1) | 35 (17.4) | |
| 55–64 | 103 (22.5) | 47 (18.2) | |
| ≥ 65 | 108 (15.1) | 45 (17.9) | |
| Race/ethnicity | .28 | ||
| White | 287 (63.6) | 120 (59.1) | |
| Black | 43 (15.4) | 18 (18.4) | |
| Hispanic | 22 (14.7) | 7 (9.1) | |
| Other | 34 (3.9) | 13 (11.7) | |
| Multiracial | 11 (2.4) | 9 (1.7) | |
| Type of interview | .4 | ||
| Cell phone | 150 (46.6) | 73 (54.2) | |
| Landline | 254 (53.4) | 98 (45.8) |
Note. Frequencies are unweighted; percentages are weighted. The states and territories are DE, HI, ID, IN, IA, KS, KY, LA, MD, MN, MT, NV, NY, OH, PA, VT, VA, WI, WY, and Guam.
Table 3 presents results among only those transgender individuals with conflicting sex designation. Of the transgender individuals who received seemingly erroneous sex-specific questions, 25.0% of female-to-male individuals designated male reported that they had received a PSA test, and 29.4% of male-to-female individuals designated female reported that they had a hysterectomy. Although no male-to-female individuals reported being pregnant, 31 male-to-female individuals answered “no” to that question, meaning that they were at least asked this question.
TABLE 3—
Sex-Specific Survey Items by Sex and Gender Identity Among Transgender Individuals in 20 States and Territories: Behavioral Risk Factor Surveillance System, United States, 2014
| Variable | Male-to-Female, No. (%) | Female-to-Male, No. (%) | Gender-Nonconforming, No. (%) |
| Malea | |||
| PSA test (men aged ≥ 40 y only) | |||
| Yes | 96 (39.3) | 13 (25.0) | 17 (30.9) |
| No | 87 (35.7) | 25 (48.1) | 17 (30.9) |
| Don’t know | 5 (2.0) | 1 (1.9) | 1 (1.8) |
| Refused | 0 (0.0) | 0 (0.0) | 0 (0.0) |
| Missing | 56 (23.0) | 13 (25.0) | 20 (36.4) |
| Femaleb | |||
| Mammogram (women only) | |||
| Yes | 91 (76.5) | 131 (81.9) | 41 (67.2) |
| No | 27 (22.7) | 28 (17.5) | 19 (31.2) |
| Don’t know | 0 (0.0) | 0 (0.0) | 0 (0.0) |
| Refused | 1 (0.8) | 1 (0.6) | 1 (1.6) |
| Missing | 0 (0.0) | 0 (0.0) | 0 (0.0) |
| Papanicolaou test (women only) | |||
| Yes | 90 (75.6) | 144 (67.9) | 52 (85.3) |
| No | 27 (22.7) | 15 (7.1) | 8 (13.1) |
| Don’t know | 1 (0.8) | 0 (0.0) | 0 (0.0) |
| Refused | 1 (0.8) | 1 (0.6) | 1 (1.6) |
| Missing | 0 (0.0) | 0 (0.0) | 0 (0.0) |
| Hysterectomy (nonpregnant women only) | |||
| Yes | 35 (29.4) | 48 (30.0) | 22 (36.1) |
| No | 82 (68.9) | 110 (68.7) | 37 (60.7) |
| Don’t know | 1 (0.8) | 0 (0.0) | 0 (0.0) |
| Refused | 1 (0.8) | 1 (0.6) | 1 (1.6) |
| Missing | 0 (0.0) | 1 (0.6) | 1 (1.6) |
| Pregnant (women aged ≤ 45 y) | |||
| Yes | 0 (0.0) | 1 (0.6) | 1 (1.6) |
| No | 31 (26.1) | 37 (23.1) | 26 (42.6) |
| Don’t know | 0 (0.0) | 0 (0.0) | 0 (0.0) |
| Refused | 0 (0.0) | 1 (0.6) | 0 (0.0) |
| Missing | 88 (73.9) | 121 (75.6) | 34 (55.7) |
Note. PSA = prostate-specific antigen. Frequencies and percentages are unweighted. The states and territories are DE, HI, ID, IN, IA, KS, KY, LA, MD, MN, MT, NV, NY, OH, PA, VT, VA, WI, WY, and Guam.
Male-to-female n = 244; female-to-male n = 52; gender-nonconforming n = 55.
Male-to-female n = 119; female-to-male n = 160; gender-nonconforming n = 61.
DISCUSSION
Results of this descriptive study suggest that, in the 2014 BRFSS data collection of gender identity and sex, there are several assumptions made in data collection that may present challenges for data analysis. Assuming respondents’ sex by interviewer assessment of vocal timbre may have resulted in administration of incorrect sex-specific survey items because an interviewer incorrectly assumed the respondent’s sex. For transgender respondents, this means that a female-to-male transgender person whom an interviewer assigns a male sex will receive questions about prostate screening, and a male-to-female person whom an interviewer assigned female sex will receive questions about Papanicolaou tests, pregnancy, and hysterectomy history.
This misclassification has several implications for transgender respondents and the researchers who use these data. First, this may result in inflation of “don’t know” or refusal responses to items that actually should not have been asked in the first place, thereby creating a false nonresponse rate. Second, it generates seemingly implausible results among transgender individuals (e.g., 25% of female-to-male transgender individuals designated as male reported receiving a PSA test despite the assumption that these individuals do not have prostates). Third, it is difficult to interpret these findings if they are, in fact, true responses. It is unclear why the percentage of implausible responses to sex-specific items by transgender participants varies by type of screening or condition. It is possible that transgender respondents may have been unfamiliar with certain sex-specific medical procedure terminology, felt too uncomfortable to point out to the interviewer that they were asking the wrong questions, or that they were not transgender at all and had erroneously identified themselves as such. Further investigation would be valuable to determine respondent’s understanding of language used in gender identity questions, specifically in the area of survey design and flow patterns predicated on the respondent’s sex. Some studies show promising results for the validity of gender-identity measures when administered to cisgender respondents.13
If female-to-male individuals are indeed receiving PSA tests, it would be a misapplication of a medical screening procedure, which has implications for patient–provider education and health system protocols to prevent such instances of misapplied medical procedures. Further investigation would shed light on the reason for these results—misapplication of medical screening versus incorrect data on sex versus misunderstanding of the question (e.g., a respondent might not know what a PSA test is) versus dishonest responses. Moreover, not only does the approximation of sex by vocal timbre introduce bias into estimates for transgender respondents, but the current data collection protocol may also bias estimates of sex-related health behaviors collected in the survey.
Although the addition of gender identity items to the BRFSS is very much needed, from a survey administration perspective, the 2014 CDC guidelines for the BRFSS have not specified how to administer sex-specific items in conjunction with the gender-identity module. Optional modules like the gender-identity module are added after required core components, which include sex-specific items. In such cases, sex-specific items can be erroneously administered to transgender participants. Moreover, it is unclear if the erroneous administration of these sex-specific items may have interrupted interviewer–respondent rapport or potentially contributed to transgender individuals choosing to end the survey before completion.
From a participant perspective, answering a question “Are you male or female?” can be unclear and uncomfortable for many transgender people; sex and gender are often conflated and understandings of the terms vary culturally.12 Transgender respondents may wonder if they should respond according to their birth sex or their current gender, which can be an invalidating experience. Thus, in addition to the psychometric testing of gender-identity questions, survey developers should also be aware that items that seem straightforward, such as sex, may require more deft wording and placement to improve the survey experience for all participants and improve data quality. A simple solution and one that has been recommended by other researchers in the field is to universally ask all participants their birth sex, and include a follow-up item to confirm natal sex if a person identifies as transgender.13 For example: “Was your sex at birth (1) male (2) female?” A clarifying definition may be necessary for cisgender participants who are unfamiliar with the phrasing “sex at birth.” Furthermore, sex-specific items should follow the gender-identity module rather than precede.
Limitations
Several limitations should be noted. First, the gender identity module was not used nationally, which may limit generalizability of our findings.
Second, there were no data on how often the sex of the respondent was either assigned by the interviewer or asked of the respondent, precluding our ability to estimate potential misclassification bias. We assumed that the interviewers themselves had an adequate understanding of how to administer the gender-identity survey item and respond to any questions or confusion from a participant; however, it is unclear whether additional training resources were offered to states that chose to administer the gender-identity module.
Lastly, because the collection of gender-identity data in large, probability-based samples is a new practice, there is no literature with which we can compare these findings.
Public Health Implications
Phone-based surveys can improve the collection of sex- and gender-based items and provide rigorously collected evidence of health disparities experienced by transgender populations. However, more circumspect survey planning is needed to reduce measurement error, including interviewers asking all respondents their sex, and using specific definitions such as “sex assigned at birth.” Moreover, researchers who use the BRFSS data to examine transgender health should be aware of potential inconsistencies in the data (e.g., counterintuitive responses, false missingness, inflation of “don’t know” responses) stemming from the BRFSS architecture around sex-related items, sex designation, and gender identity.
ACKNOWLEDGMENTS
We would like to thank members of the Center for LGBT Health Research working group at the University of Pittsburgh for their insights and contributions.
HUMAN PARTICIPANT PROTECTION
This was a secondary analysis of a national data set and no institutional review board approval was required.
REFERENCES
- 1. Flores AR, Herman JL, Gates GJ, Brown TNT. How many adults identify as transgender in the United States? Los Angeles, CA: The Williams Institute; 2016.
- 2.Crissman HP, Berger MB, Graham LF, Dalton VK. Transgender demographics: a household probability sample of US adults, 2014. Am J Public Health. 2017;107(2):213–215. doi: 10.2105/AJPH.2016.303571. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Meerwijk EL, Sevelius JM. Transgender population size in the United States: a meta-regression of population-based probability samples. Am J Public Health. 2017;107(2):e1–e8. doi: 10.2105/AJPH.2016.303578. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Grant JM, Mottet LA, Tanis J, Harrison J, Herman JL, Keisling M. Injustice at every turn. Washington, DC: National Center for Transgender Equality, National Gay and Lesbian Task Force; 2011.
- 5.Bradford J, Reisner SL, Honnold JA, Xavier J. Experiences of transgender-related discrimination and implications for health: results from the Virginia Transgender Health Initiative Study. Am J Public Health. 2013;103(10):1820–1829. doi: 10.2105/AJPH.2012.300796. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Herbst JH, Jacobs ED, Finlayson TJ et al. Estimating HIV prevalence and risk behaviors of transgender persons in the United States: a systematic review. AIDS Behav. 2008;12(1):1–17. doi: 10.1007/s10461-007-9299-3. [DOI] [PubMed] [Google Scholar]
- 7.Stotzer RL. Violence against transgender people: a review of United States data. Aggress Violent Behav. 2009;14(3):170–179. [Google Scholar]
- 8.Conron KJ, Scott G, Stowell GS, Landers SJ. Transgender health in Massachusetts: results from a household probability sample of adults. Am J Public Health. 2012;102(1):118–122. doi: 10.2105/AJPH.2011.300315. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Reisner SL, Poteat T, Keatley J et al. Global health burden and needs of transgender populations: a review. Lancet. 2016;388(10042):412–436. doi: 10.1016/S0140-6736(16)00684-X. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Reisner SL, Deutsch MB, Bhasin S et al. Advancing methods for US transgender health research. Curr Opin Endocrinol Diabetes Obes. 2016;23(2):198–207. doi: 10.1097/MED.0000000000000229. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Landers S, Kapadia F. The health of the transgender community: out, proud, and coming into their own. Am J Public Health. 2017;107(2):205–206. doi: 10.2105/AJPH.2016.303599. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. The GeniUSS Group. Best practices for asking questions to identify transgender and other gender minority respondents on population-based surveys. Los Angeles, CA: The Williams Institute; 2014.
- 13.Reisner SL, Conron KJ, Baker K et al. “Counting” transgender and gender-nonconforming adults in health research recommendations from the Gender Identity in US Surveillance Group. TSQ: Transgender Studies Quarterly. 2015;2(1):34–57. [Google Scholar]
- 14.Pega F, Reisner SL, Sell RL, Veale JF. Transgender health: New Zealand’s innovative statistical standard for gender identity. Am J Public Health. 2017;107(2):217–221. doi: 10.2105/AJPH.2016.303465. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Clopper CG, Pisoni DB. The nationwide speech project: a new corpus of American English dialects. Speech Commun. 2006;48(6):633–644. doi: 10.1016/j.specom.2005.09.010. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Sataloff RT, Rosen DC, Hawkshaw M, Spiegel JR. The aging adult voice. J Voice. 1997;11(2):156–160. doi: 10.1016/s0892-1997(97)80072-0. [DOI] [PubMed] [Google Scholar]
- 17.Purnell T, Idsardi W, Baugh J. Perceptual and phonetic experiments on American English dialect identification. J Lang Soc Psychol. 1999;18(1):10–30. [Google Scholar]