Archives of Sexual Behavior

, Volume 43, Issue 3, pp 413–422

The Dubious Assessment of Gay, Lesbian, and Bisexual Adolescents of Add Health

Authors

    • Department of Human DevelopmentCornell University
  • Kara Joyner
    • Department of SociologyBowling Green State University
Invited Essay

DOI: 10.1007/s10508-013-0219-5

Cite this article as:
Savin-Williams, R.C. & Joyner, K. Arch Sex Behav (2014) 43: 413. doi:10.1007/s10508-013-0219-5

Abstract

In this essay, we argue that researchers who base their investigations of nonheterosexuality derived from reports of romantic attractions of adolescent participants from Wave 1 of Add Health must account for their disappearance in future waves of data collection. The high prevalence of Wave 1 youth with either both-sex or same-sex romantic attractions was initially striking and unexpected. Subsequent data from Add Health indicated that this prevalence sharply declined over time such that over 70 % of these Wave 1 adolescents identified as exclusively heterosexual as Wave 4 young adults. Three explanations are proposed to account for the high prevalence rate and the temporal inconsistency: (1) gay adolescents going into the closet during their young adult years; (2) confusion regarding the use and meaning of romantic attraction as a proxy for sexual orientation; and (3) the existence of mischievous adolescents who played a “jokester” role by reporting same-sex attraction when none was present. Relying on Add Health data, we dismissed the first explanation as highly unlikely and found support for the other two. Importantly, these “dubious” gay, lesbian, and bisexual adolescents may have led researchers to erroneously conclude from the data that sexual-minority youth are more problematic than heterosexual youth in terms of physical, mental, and social health.

Keywords

Add HealthSexual orientationGayLesbianBisexualAdolescents

Add Health’s Nonheterosexual Adolescents

During the past decade, many of us have reviewed numerous manuscripts based on Wave 1 data from the National Longitudinal Study of Adolescent Health (Add Health), many of which purported to investigate sexual orientation effects on adolescent development, risk factors, and physical/mental health based on a romantic attraction variable. Due to the overall quality of the Add Health study (e.g., the large, representative national sample of youth) and the dearth of data on sexual-minority adolescents, many of these studies were published. In sum, they found that Wave 1 nonheterosexual youth (to enhance statistical power, both-sex and same-sex participants were usually combined) were more likely than heterosexual youth to report both negative and positive experiences. Specifically, nonheterosexual youth were more likely to be victimized or attacked by peers, have suicidal thoughts, attempt suicide, abuse alcohol/drugs, suffer depression, be less attached to parents, argue with parents, have more interpersonal school problems, perceive themselves to be at greater risk for HIV/STDs, seek counseling, play volleyball, and wrestle (Russell, Franz, & Driscoll, 2001; Russell & Joyner, 2001; Udry & Chantala, 2005; Ueno, 2005; Zipp, 2011).

Although several warning signs were apparent soon after Wave 1 data were collected that some responses to having both- or same-sex romantic attraction (i.e., sexual orientation) were suspect in terms of accuracy or authenticity, these cautionary omens were not then or now heeded. The primary red flag was the considerably larger percent of sexual-minority youth in Add Health compared to previous investigations, driven in large part by the preponderance of bisexual (both-sex attracted) boys. However, investigators continued to include Wave 1 romantic attraction data to identify sexual-minority youth. We offer this essay, with data, to forestall such wrong-headed scholarly work in the future. First, a brief overview of the Add Health study is provided and then we review initial indications that something was horribly amiss with the Add Health sample of nonheterosexuals. We review several possibilities as to why gay, lesbian, and bisexual Wave 1 adolescents did not become gay, lesbian, and bisexual young adults.

Add Health

Add Health (funded by the Eunice Kennedy Shriver National Institute of Child Health and Human Development and several other agencies) was one of the few longitudinal, nationally representative studies of its kind. The primary purpose of Add Health is to assess various social and familial contextual variables that influence health, well-being, and health-related behaviors (Fan et al., 2006; Tolman & McClelland, 2011). Add Health data have been labeled “one of the most important resources for research…[and] has provided unparalleled data on the lives of U.S. adolescents” (Tolman & McClelland, 2011, p. 249). Wave 1 data were amassed during 1994–1995 from 90,118 youth in Grades 7 through 12 from a representative sample of schools. This was the first of four waves: Wave 2 was collected just over a year later, Wave 3 in 2001–2002, and Wave 4 in 2007–2009 (Harris, 2009).

Wave 1 was collected in school settings from adolescents through paper-and-pencil, self-administered questionnaires, and from school administrators and parents. A subsample composed of a core sample and special-purpose over-samples of youth, numbering 20,747 (M age = 15.8 years), completed an interviewer-assisted questionnaire in the youths’ homes about a year later, as did about 85 % of their parents. For further details regarding the Add Health study design, see http://www.cpc.unc.edu/projects/addhealth.

In sections of the interview that included more sensitive questions, the interviewers instructed participants to listen to recorded interview questions via headphones and enter their responses on a laptop computer (audio computer-assisted self-interview). They could reply yes or no to the questions “Have you ever had a romantic attraction to a female?” and “Have you ever had a romantic attraction to a male?” Based on answers to these two questions, youth were classified as heterosexual, bisexual, homosexual, or asexual (no attraction)—with the middle two frequently combined as “any same-sex attraction.” It is noteworthy that these questions were posed without definition of what constitutes romantic attraction—it has been assumed (without empirical verification) by investigators that it assessed sexual orientation. Romantic attraction, however, is an unusual term that, outside of Add Health, has seldom been used as a measure of sexual orientation.

The sample we use in this report begins with participants who have a sample weight for the Wave 4 interview (N = 14,800), as previous studies based on Add Health restricted their samples to participants with such information. We excluded from the sample those who had an inconsistent biologic sex marked by interviewers for the Waves 1 and 4 interviews, reducing the sample to 14,786. Excluding participants who failed to provide a valid response to the sexual orientation identity question at Wave 4 reduced the sample to 14,723, while dropping those who did not provide a valid response to the romantic attraction question at Wave 4 further reduced the sample to 14,700. Finally, removing youth who did not reply to the romantic attraction question at Wave 1 brought the sample to 14,553.

Evidence Something was Amiss

We were initially cautiously optimistic by the large number of Wave 1 youth who reported having any same-sex romantic attraction until we discovered that these individuals were frequently not the same individuals who indicated nonheterosexuality a year later in Wave 2—and, later still, during their young adult years in Waves 3 and 4 (Savin-Williams, Joyner, & Rieger, 2012; Savin-Williams & Ream, 2007). In particular, we noted two temporal trends in prevalence rates. One was the increasingly high number of women who reported romantic attraction to both sexes and identified as mostly heterosexual, which was consistent with a recent review of the literature (Savin-Williams & Vrangalova, 2013). The second, more germane and ominous for this essay, was the dramatic decline of both-sex attracted boys (girls to a lesser degree) after Wave 1.

Given other data from that time, the extraordinary high prevalence of nonheterosexual middle and high school adolescents—7.3 % of boys and 5.0 % of girls reported same-sex or both-sex romantic attraction (Russell & Joyner, 2001)—was remarkable. For example, in the Growing Up Survey collected in 1999 of over 10,000 youth ages 12-to-17 years, just over 1 % of both girls and boys reported being solely or equally attracted to men and women (Austin et al., 2004). The dramatic reduction, especially among boys, of nonheterosexuality occurred just over a year later in Wave 2 when 4.5 % of boys and girls reported having any same-sex romantic attraction (Savin-Williams & Ream, 2007).

Add Health Principal Investigator J. Richard Udry was the first to point out this instability of romantic attraction between the first two waves (Udry & Chantala, 2005). Udry and Chantala reported that, of the Wave 1 69 boys who reported romantic attraction only to boys and never to girls, 48 % reported at Wave 2 that during the past year they had only been attracted to girls; 35 % reported no attraction to anyone; 11 % reported attraction only to boys; and 6 % reported attraction to both sexes. Thus, 83 % of the Wave 1 adolescent gay boys were neither gay nor bisexual at Wave 2. They concluded that “the responses are not ‘invalid’; they are only unstable” (Udry & Chantala, 2005, p. 486). Presumably, by combining data from the two waves, one might achieve a more accurate prediction of stable adult sexual orientation, but they expressed doubts. Perhaps however subtly, they were acknowledging that a problem existed with Wave 1 romantic attraction data.

This problem was readily apparent in a later study that compared romantic attraction data from Waves 1, 2, and 3 (see Table 2 in Savin-Williams & Ream, 2007). For example, of the Wave 1 boys who reported romantic attraction to both sexes, over 80 % reported no romantic attraction to boys at Wave 3 (i.e., they were exclusively attracted to girls); of the Wave 1 both-sex attracted girls, more than half reported only attraction to boys at Wave 3. Of Wave 1 boys who were exclusively romantically attracted to other boys, over 70 % reversed themselves at Wave 3 and were only attracted to girls; for girls, over half were now only attracted to boys.

We extended this appraisal to assess how Wave 1 nonheterosexual adolescents fared at Wave 4 when participants were, on average, 28 years old (Table 1). Of Wave 1 boys who indicated they had any same-sex romantic attraction, over 80 % identified at Wave 4 as exclusively heterosexual or only had opposite-sex romantic attraction. Just over 10 % reported that they were bisexual, mostly homosexual, or homosexual. Reversing the perspective, of the Wave 4 bisexual, mostly homosexual, and homosexual young men, 75 % reported no both-sex or same-sex romantic attraction at Wave 1.
Table 1

Wave 4 sexual orientation identity and romantic attraction by Wave 1 romantic attraction: Add Health participants

 

N

%

Wave 1 romantic attraction

Opposite sex only

Both sexes

Same sex only

Neither sex

N

%

N

%

N

%

N

%

Men

6,804

100.0

5,536

81.4

405

6.0

49

0.7

814

12.0

Wave 4 identity

 100 % Heterosexual

6,333

93.1

5,204

94.0

342

84.4

37

75.5

750

92.1

 Mostly heterosexual

226

3.3

184

3.3

15

3.7

2

4.1

25

3.1

 Bisexual

44

0.6

30

0.5

6

1.5

1

2.0

7

0.9

 Mostly homosexual

55

0.8

35

0.6

13

3.2

0

0.0

7

0.9

 100 % Homosexual

125

1.8

72

1.3

27

6.7

9

18.4

17

2.1

 Asexual

21

0.3

11

0.2

2

0.5

0

0.0

8

1.0

Wave 4 attraction

 Opposite sex only

6,435

94.6

5,304

95.8

346

85.4

33

67.3

752

92.4

 Both sexes

144

2.1

100

1.8

22

5.4

4

8.2

18

2.2

 Same sex only

160

2.4

94

1.7

36

8.9

10

20.4

20

2.5

 Neither sex

65

1.0

38

0.7

1

0.2

2

4.1

24

2.9

Women

7,749

100.0

6,565

84.7

307

4.0

100

1.3

777

10.0

Wave 4 identity

 100 % Heterosexual

6,213

80.2

5,340

81.3

170

55.4

64

64.0

639

82.2

 Mostly heterosexual

1,181

15.2

986

15.0

92

30.0

19

19.0

84

10.8

 Bisexual

180

2.3

128

1.9

27

8.8

3

3.0

22

2.8

 Mostly homosexual

67

0.9

49

0.7

6

2.0

6

6.0

6

0.8

 100 % Homosexual

74

1.0

45

0.7

11

3.6

7

7.0

11

1.4

 Asexual

34

0.4

17

0.3

1

0.3

1

1.0

15

1.9

Wave 4 attraction

 Opposite sex only

6,957

89.8

5,971

91.0

215

70.0

74

74.0

697

89.7

 Both sexes

588

7.6

460

7.0

74

24.1

13

13.0

41

5.3

 Same sex only

120

1.5

79

1.2

12

3.9

11

11.0

18

2.3

 Neither sex

84

1.1

55

0.8

6

2.0

2

2.0

21

2.7

Note. The sample includes participants who provided valid responses to questions about romantic attraction at both waves and who provided valid responses to the question about sexual orientation identity at Wave 4. Participants with inconsistent biologic sex classification at Waves 1 and 4 were excluded from this sample

The girls were somewhat less inconsistent in reporting their sexual data (Table 1). Of those with same-sex romantic attraction at Wave 1, nearly 60 % identified as exclusively heterosexual; 30 % as mostly heterosexual; and over 70 % only had opposite-sex romantic attraction at Wave 4. Just over 10 % of the girls identified as bisexual, mostly homosexual, or homosexual. Reversing the perspective, of the young women who identified as bisexual, mostly homosexual, or homosexual at Wave 4, 81 % reported no both-sex or same-sex attraction at Wave 1.

Thus, in the approximately 13 years between Waves 1 and 4, regardless of whether the measure was identical across waves (romantic attraction) or discrepant in words but not in theory (romantic attraction and sexual orientation identity), approximately 80 % of adolescent boys and half of adolescent girls who expressed either partial or exclusive same-sex romantic attraction at Wave 1 “turned” heterosexual (opposite-sex attraction or exclusively heterosexual identity) as young adults.

One final set of analyses is presented before we suggest possible reasons for these findings. To help us assess whether the construct/measurement issue (romantic attraction versus sexual orientation identity) was driving results, we compared the two constructs at Wave 4 (Table 2). Whereas over 99 % of young adults with opposite-sex romantic attraction identified as heterosexual or mostly heterosexual and 94 % of those with same-sex romantic attraction identified as homosexual or mostly homosexual, 33 % of both-sex attracted men identified as heterosexual (just 6 % of both-sex attracted women identified as heterosexual). These data indicated that young adult men and women generally understood the meaning of romantic attraction to the opposite- or same-sex to imply a particular (and consistent) sexual orientation identity, with one glaring exception—a substantial subset of young adult men who, despite their stated both-sex romantic attraction, identified as heterosexual.
Table 2

Wave 4 sexual orientation identity by Wave 4 romantic attraction: Add Health participants

 

N

%

Wave 4 romantic attraction

Opposite sex only

Both sexes

Same sex only

Neither sex

N

%

N

%

N

%

N

%

Men

6,804

100.0

6,435

94.6

144

2.1

160

2.4

65

1.0

Wave 4 identity

 100 % Heterosexual

6,333

93.1

6,227

96.8

48

33.3

2

1.3

56

86.2

 Mostly heterosexual

226

3.3

187

2.9

37

25.7

0

0.0

2

3.1

 Bisexual

44

0.6

9

0.1

33

22.9

2

1.3

0

0.0

 Mostly homosexual

55

0.8

0

0.0

20

13.9

35

21.9

0

0.0

 100 % Homosexual

125

1.8

0

0.0

4

2.8

121

75.6

0

0.0

 Asexual

21

0.3

12

0.2

2

1.4

0

0.0

7

10.8

Women

7,749

100.0

6,957

89.8

588

7.6

120

1.5

84

1.1

Wave 4 Identity

 100 % Heterosexual

6,213

80.2

6,116

87.9

35

6.0

4

3.3

58

69.0

 Mostly heterosexual

1,181

15.2

814

11.7

355

60.4

4

3.3

8

9.5

 Bisexual

180

2.3

6

0.1

166

28.2

4

3.3

4

4.8

 Mostly homosexual

67

0.9

2

0.0

29

4.9

36

30.0

0

0.0

 100 % Homosexual

74

1.0

2

0.0

2

0.3

70

58.3

0

0.0

 Asexual

34

0.4

17

0.2

1

0.2

2

1.7

14

16.7

Note. The sample includes participants who provided valid responses to questions about romantic attraction at both waves and who provided valid responses to the question about sexual orientation identity at Wave 4. Participants with inconsistent biologic sex classification at Waves 1 and 4 were excluded from this sample

Why Exit Nonheterosexuality?

Accounting for the unexpected and unprecedented level of youth exiting nonheterosexuality has been challenging because of the limited data available in Add Health and because previous empirical literature that could guide us was sparse. After lengthy discussions, we decided three possibilities were worthy of consideration. We invite readers to expand on these options.

First, the inconsistent adolescents were out during adolescence and then decided to hide their sexuality during their young adulthood (“It is undesirable being gay/lesbian/bisexual so I’m going back into the closet.”). The question: Are these data consistent with what is known about gay youth development?

Second, given the unusual and vague term romantic attraction, these were adolescents who misunderstood the question (“Does romantic attraction mean I like someone or want to have sex with them?”). The question: Is the construct of romantic attraction itself the problem?

Third, the exiting nonheterosexual adolescents were heterosexual jokesters who decided to have a bit of fun with the research (“I’ll screw the researchers”). The question: Are the inconsistencies a manifestation of adolescent malfeasance?

Explanation 1: Developmental Change

Russell and Seif (2010) noted the discrepancies between Waves 1 and 2 assessments of romantic attraction and attributed the steep drop in nonheterosexuality not to poor measures or bad data (consistent with Udry & Chantala, 2005) but to the nature of same-sex sexuality during adolescence. They interpreted the inconsistency as support for a prototypical adolescent response to living in a heterosexist culture (going back into the closet) and to female sexual fluidity.

We could not find a study based on other data that supported this unusual proposal—that such a large majority of nonheterosexual youth moves from being out during adolescence to going in the closet after adolescence or that such a substantial number of heterosexual youth report that they are gay, lesbian, or bisexual and then decide during early adulthood that they are heterosexual after all. Regarding the first proposition, based on our reading of the scientific and popular literature, it is our impression that if youth during the mid-1990s were sufficiently brave to be “out” during their teenage years, most would have maintained this status as they aged (early adulthood) into increasingly tolerant environments (e.g., college) at a time (mid-2000s) of unparalleled acceptance of gay people (Pew Research Center, 2013). Regarding the second proposition, the penalty (social stigma) for not being heterosexual during the mid-1990s when Wave 1 was collected would have been sufficiently intense that we doubt that so many exclusive heterosexual adolescents voluntarily claimed this status–unless, of course, there was another reason to make this declaration (see Explanations 2 and 3).

Thus, although shifts in stated sexual preferences are to be expected during adolescence, the observed patterns in Add Health reverse prior developmental research. That is, few nonheterosexual adolescents go into the closet during young adulthood; rather, this has been reported to be a good time to disclose one’s sexuality (Savin-Williams, 2005). Add Health Wave 4 data supported the common developmental finding that it is not adolescence but young adulthood when most nonheterosexual individuals are out (Table 1). Of the 545 gay, lesbian, and bisexual millennial young adults, nearly two thirds did not report their nonheterosexuality during their adolescence in the 1990s (were closeted at Wave 1) but instead claimed to be heterosexual or asexual (only had opposite-sex attraction or attracted to neither sex). Indeed, the 1990s were so not conducive to being out than only 0.8 % of all Add Health youth were consistently nonheterosexual from Wave 1 to Wave 4.

If, however, we were to believe the Wave 1 adolescents’ self-reports of nonheterosexuality, then the relevant question is which nonheterosexuals might be out during adolescence but not during young adulthood. One answer, we speculated, is that these individuals were those with much to gain by going into the closet–ideologically, the least politically liberal and the most religiously conservative. As indicated in Table 3, as expected, those consistently nonheterosexual across time were the most politically liberal, the least politically conservative, and least religious (“How often have you attended church, synagogue, temple, mosque, or religious services in the past 12 months?”) during Wave 4; however, young adults who “went into the closet” were no different from consistently heterosexuals in their political and religious orientations (both sexes).
Table 3

Waves 1–4 changes in select variables by sexual orientation group identity: Add Health participants

 

Consistently heterosexual

W1 nonheterosexual

W4 heterosexual 

Consistently nonheterosexual

N

M

N

M

N

M

Men

 Liberala

4,931

0.237

343

0.236

71

0.535***

 Conservativeb

4,931

0.281

343

0.283

71

0.183*

 Religious attendancec

5,202

0.152

378

0.161

73

0.082*

 PPVT at W1d

4,932

102.8

358

98.5***

73

106.9**

 H.S. English GPAe

3,474

2.22

230

1.90***

47

2.51*

 Any collegef

5,203

0.633

379

0.501***

73

0.781**

 Very honest W1 g

5,184

0.904

377

0.788***

73

0.863

 Impatient/bored W1 h

5,203

0.131

379

0.182*

73

0.096

 Delinquency W1i

5,157

3.246

375

3.403

73

3.192

 Unexcused absence W1j

5,204

0.323

379

0.430***

73

0.452*

 Other sex popular W1 k

3,400

0.379

226

0.410

52

0.640***

 PR(boy) W2l

3,409

0.636

249

0.642

48

0.448***

Women

 Liberal

5,039

0.276

218

0.312

163

0.521***

 Conservative

5,039

0.261

218

0.206#

163

0.129***

 Religious attendance

5,340

0.233

234

0.256

171

0.129***

 PPVT at W1

5,113

100.4

224

96.2***

164

106.9***

 H.S. English GPA

3,710

2.66

145

2.56

104

2.49*

 Any college

5,340

0.728

234

0.675#

171

0.743

 Very honest W1

5,325

0.940

233

0.893*

171

0.953

 Impatient/bored W1

5,339

0.080

234

0.081

171

0.111

 Delinquency W1

5,317

1.891

232

2.668**

171

4.129***

 Unexcused absence W1

5,340

0.266

234

0.363**

171

0.404***

 Other sex popular W1

3,745

0.328

153

0.272*

113

0.404*

 PR(boy) W2

3,653

0.335

141

0.324 

87

0.299 

Note. All variables measured at Wave 4 except as noted. Consistently heterosexual: Opposite-sex attracted at W1 and 100 % heterosexual at W4. Inconsistent: Same-sex attracted at W1 and 100 % heterosexual at W4. Consistently nonheterosexual: Same-sex attracted W1 and not 100 % straight or asexual W4

aThose liberal (1.0) versus middle of the road (0.0)

bThose conservative (1.0) versus middle of the road (0.0)

cAttendance at religious services during the week, from one or more times (1.0) versus none (0.0)

dPeabody Picture Vocabulary Test standardized score, from 14 to 146

eGrade point grade in English class from 0 (failure) to 4 (A or excellent)

fAttended at least some college (1.0) or no college (0.0)

gSelf-reported response to questions, either “very honestly” (1.0) or “not honestly at all” or “somewhat honestly” (0.0)

hInterviewer rated whether participant appeared bored or impatient during interview, no (0.0) versus yes (1.0)

iOn 10 items, participants reported their recent delinquent or undesirable behaviors (vandalism, violence, weapons), from never (0.0) to 5 or more times (3.0) with range 0–30

jHow frequently skipped school for a full day without an excuse, from none (0.0) to one or more (1.0)

kNumber of friendship nominations received from a different sex by the total number of friendship nominations received from either sex (participants with zero total votes not included in calculation), from none (0.0) to all (1.0)

lBased on responses to 16 questions to predict the probability that participant is a boy (e.g., takes risks), from 0.0 to 1.0

# p < .10; * p < .05; ** p < .01; *** p < .001 (two-tailed tests of statistical significance assume unequal variances and use the Cochran method to estimate the df)

In addition, given the literature on female fluidity, physiological arousal, and sexual orientation (Chivers, Seto, Lalumiere, Laan, & Grimbos, 2010; Diamond, 2008; Rieger & Savin-Williams, 2012a; Savin-Williams, 2005), we expected women to be more inconsistent than men in their responses over time and across measures. However, in the data reported here, in all situations inconsistent men rather than women (especially both-sex attracted) were the least consistent. Thus, on all accounts, the Russell and Seif (2010) explanation makes neither empirical nor developmental sense.

Explanation 2: Confusion with the Romantic Attraction Construct

Soon after numerous Wave 1 Add Health studies were published, Fan et al. (2006) concluded that, similar to other large-scale datasets, Add Health included adolescent participants who should be classified as inaccurate responders, who gave answers either in carelessness or confusion, and jokesters, who intentionally provided false answers. To us, their analysis suggested the prospect that Add Health might also include adolescents who fashioned themselves as accidental gays, lesbians, and bisexuals. In this section, we consider their first suggestion: adolescents classified as nonheterosexual due to their confusion or inability to comprehend the sexual orientation measure included in Wave 1.

Even though romantic attraction was perhaps a poor choice as an indicator of sexual orientation, it was likely a necessary political compromise to accommodate conservative opposition to U.S. government funding of sex research that included adolescent minors (for a discussion of the American Teenage Study, the precursor of Add Health, that failed due to early 1990s politics, see Udry, 1993). Presumably, terms such as sexual identity, sexual attraction, sexual orientation, and the more incendiary gay, lesbian, and bisexual identities were avoided to forestall funding cuts. This changed when the Add Health participants became young adults and a 5-point measure of sexual orientation identity was included in Waves 3 and 4–a frequently used assessment of sexual orientation with high concordance with other sexual orientation indicators (physiological arousal, sexual attraction, fantasy, desire, behavior, and identity) (Schwartz, Kim, Kolundzija, Rieger, & Sanders, 2010).

To account for this precipitous exit from nonheterosexuality, a measurement confusion explanation was offered in an earlier article—adolescents and, perhaps, young adults did not understand what romantic attraction meant (Savin-Williams & Ream, 2007). This explanation is difficult to assess given the data available; however, evidence points to the possibility. On measures of intelligence and education, the inconsistent participants scored the lowest on the Peabody Picture Vocabulary Test, had the lowest cumulative English grade point average in high school (boys only), and were the least likely to attend college (Table 3). On most measures, they were closer to consistently heterosexuals than consistently nonheterosexuals.

In addition, as noted earlier, Wave 4 young adults (both sexes) who reported being same-sex and opposite-sex romantically attracted identified appropriately as homosexual/mostly homosexual and heterosexual/mostly heterosexual (respectively) (Table 2). So, too, both-sex romantically attracted young women selected a suitable sexual orientation identity at Wave 4 (mostly heterosexual, bisexual, or mostly homosexual). More problematic, however, is explaining the one third of Wave 4 young adult men who, despite being both-sex attracted, identified as exclusively heterosexual. They clearly did not go into the closet during young adulthood as proposed by Explanation 1 because they reported being attracted to males at Wave 4. This leads us to speculate that a substantial number of the Wave 1 both-sex attracted inconsistent male youth remained as young adults confused about the measurement of romantic attraction.

Explanation 3: Mischievous or Jokester Adolescents

The second possibility offered by Fan et al. (2006) of jokester gays, lesbians, and bisexuals is our third explanation for the instability of sexual orientation in Add Health. One common perspective among investigators who include adolescents in their subject pool was summarized by Cornell, Klein, Konold, and Huang (2012, p. 21): “Adolescents, because of immaturity and rebelliousness, may be tempted to offer inflated reports of their engagement in socially proscribed or illicit behaviors, or they may not take the survey seriously and mark it haphazardly, producing an elevation in otherwise low base rate behaviors”. They noted that, even if the number of these youth is small, they can exert a distorting effect on survey results and, subsequently, on professional and policy stances. Many widely used surveys, such as the Centers for Disease Control and Prevention’s Youth Risk Behavioral Surveillance Survey and Add Health, however, do not routinely use validity screening items to identify and delete participants who are not answering truthfully or carefully (Cornell et al., 2012). The result can be severe. For example, the 1987 National Adolescent Student Health Survey asked over 10,000 students in Grades 8 and 10 how often they carried a handgun to school. Approximately 2.6 % of boys responded affirmatively and, without questioning the veracity of responses, the survey results (“135,000 Guns Brought to School Every Day”) were extrapolated to become sensational headlines in major U.S. newspapers and a decade-long citation for numerous professional organizations and advocacy groups fighting school violence. By contrast, the School Climate Bullying Survey, administered to nearly 8,000 middle and high school students, screened “inappropriate (invalid) responders” and found a marked reduction of risk behaviors (e.g., drinking, victimization, depression, suicide attempts) (Cornell et al., 2012).

More pertinent here, Robinson and Espelage (2011) highlighted these effects in another dataset with a cadre of adolescents who consistently provided unusual or infrequent responses (e.g., implausible heights/weights, having more than two children), some of who reported a same-sex orientation. Reports of suicide ideation and attempts, victimization, and problems at school were, as expected, significantly higher among LGBTQ adolescents when compared with heterosexual-identified students. However, all sexual orientation differences decreased in significance when the “unusual responders” were excluded (several to non-significance).

Similar to other surveys, Add Health also has its share of unusual responders. Based on questions regarding whether they were adopted, born in a foreign country, or had an artificial limb, Fan et al. (2006) discovered that some Wave 1 adolescents gave inaccurate and/or invalid responses on the self-administered in-school questionnaire. For example, 253 youth stated on the questionnaire that they had an artificial hand, arm, leg, or foot; however, when interviewed later at home, only two of the 253 reported that they were using an artificial limb. Of the 418 inaccurate responders to the three questions, 82 youth consistently gave such a high degree of erroneous responses that the authors concluded that the youth were not merely careless or confused (Explanation 2) but were intentionally or mischievously responding. Further analyses revealed that these jokester adolescents were much more likely to report extreme levels on problem behaviors, such as drinking and smoking, and lower mean scores on positive variables, such as physical and emotional health.

To assess the possibility that the inconsistent sexual orientation youth were jokesters, we used several approaches. First, Add Health included a 4-point item in Wave 1 that requested participants at the close of the self-administered sections of the interview to assess how honestly they answered questions using the computer themselves. Inconsistent boys and girls were significantly less likely to self-report telling the truth than either the consistently heterosexual or the consistently nonheterosexual adolescents (Table 3). Of the six groups, inconsistent adolescent boys were the least likely to report that they were “very” or “completely” honest filling out the self-administered questions. Of course, we were unable to assess if individuals honestly answered the honesty question.

Second, Wave 1 interviewers were asked, “Did the respondent ever seem bored or impatient during the interview?” Inconsistent boys were most likely to be assessed as bored or impatient; the three groups of girls did not differ significantly (Table 3).

Third, a 10-item delinquency scale was included in Wave 1 that assessed self-reported frequency of involvement in various delinquent activities in the past year (“Use or threaten to use a weapon to get something from someone”). Also included was a Wave 1 question, “During this school year, how many times did you skip school for a full day without an excuse?” Inconsistent boys had the highest (not significant) level of delinquency and were significantly more likely to have an unexcused absence than heterosexual youth; inconsistent girls scored significantly higher than heterosexual but not nonheterosexual girls on both items (Table 3).

Fourth, during the administration of the Wave 1 in-school questionnaire, youth nominated up to five male and five female friends using the school attendance roster. Based on these data, we counted the number of times youth were nominated as a friend by same-sex and other-sex peers at their school. We calculated the proportion of total nominations they received from the other sex. Inconsistent boys were similar to heterosexual boys in being more popular with boys than girls; consistently nonheterosexual boys were more popular with girls than boys (see “Gender Typicality of Inconsistent Men” section below). Inconsistent girls, however, were more popular with girls than boys than either consistently heterosexual or nonheterosexual girls (Table 3).

The picture that emerged from these data was a group of inconsistent boys who, compared to other boys, were less honest on their questionnaire and more bored or impatient. They tended to be more delinquent and more likely than heterosexual boys to skip school. Inconsistent girls were less honest and less popular with boys than were other girls; they were also more delinquent and more likely to skip school than heterosexual girls. We speculate that the lesbian/bisexual girls are delinquent and skip school not because they are jokesters but for other reasons (e.g., are bullied).

Gender Typicality of Inconsistent Men

One nearly universal finding during the past two decades is that same-sex oriented men are more gender atypical than heterosexual men across childhood and adulthood, developmental domains (e.g., motor behavior, self-concepts, occupational interests), methodologies (e.g., self-report, observation of behavior), and cultures (Bailey, Dunne, & Martin, 2000; Bailey & Zucker, 1995; Lippa, 2000; Rieger & Savin-Williams, 2012b). These findings are large, with effect sizes exceeding 1. Findings regarding gender atypicality among bisexual and lesbian women were less strong. Thus, if the inconsistent boys were truly nonheterosexual, we would expect them to score high on gender atypicality, including increased popularity with girls, similar to gay and bisexual men.

For Add Health, Udry and Chantala (2005) developed a measure of gender typicality (awkwardly titled, probability of being a boy) that included items from Wave 2 that significantly differentiated boys and girls in a logistic regression model, such as sensitivity to the feelings of others. On the basis of this model, they computed a predicted probability of being a boy for each participant with Wave 2 data. Values on this measure can range from 0 to 1, with higher values indicating a greater likelihood that the participant is a boy. As expected (Table 3), consistently nonheterosexual boys were significantly lower than consistently heterosexual boys on probability of being a boy (as were those young adult men who came out during young adulthood, data not shown) and higher on popularity with girls. Inconsistent boys were indistinguishable from heterosexual boys in these two and were significantly more likely than nonheterosexual boys to be gender-typical males.

Lippa (2000) argued that the degree to which men’s gender expression is gay- versus heterosexual-typical correlates almost perfectly with the degree to which those same expressions are female- versus male-typical. Many of the effect sizes predicting sexual orientation from gender expression (or the reverse) documented in the literature are so robust that the two approach unity. In this regard, the inconsistent boys appear to be heterosexual, not gay or bisexual.

Conclusion

Data presented in this essay support the view that boys who emerged from a gay or bisexual adolescence to become a heterosexual young adulthood were, by-and-large, heterosexual adolescents who were either confused and did not understand the measure of romantic attraction or jokesters who decided, for reasons we were not able to detect, to dishonestly report their sexuality. We are thus unable to determine the proportion of inaccurate responders and jokesters. One (untestable) possibility is that the jokester effect is far greater than our data would indicate because it is likely that jokesters were, as might be expected, dishonest in their assessment of the honesty question. Thus, jokesters might well populate the false positives of Wave 1 far more than we have documented.

Tolman and McClelland (2011, p. 250) rightly concluded that Add Health data have “produced tremendous insight into the sexuality development of young people.” However, if the answer to Fan et al.’s (2006, p. 233) question “Do these inaccurate responders and jokesters introduce enough systematic data error to have a significant negative impact on substantive research?” is positive, then Add Health might have contributed little and, perhaps, distorted our understanding of sexual development among Wave 1 nonheterosexual adolescents. The distortion might have been in over-emphasizing the trauma of being young and gay (Savin-Williams, 2005). The extent of the documented pathology of nonheterosexual youth was enhanced, we believe, by the responses of a small but significant number of primarily heterosexual adolescents who were either confused or having fun as jokesters. Indeed, the extent of sexual orientation differences in mental health is greatly reduced in Wave 4 when participants were young adults and inaccurate responders and jokesters were reduced in number (Savin-Williams, Cohen, Joyner, & Rieger, 2010).

One option that Savin-Williams and Ream (2006) applied to overcome this dilemma was to trace backward the developmental footsteps of individuals who in young adulthood (Waves 3 and 4) identified as nonheterosexual to note their adolescent imprint. They used Waves 1 and 2 self-reported pubertal milestones of Wave 3 nonheterosexuals to assess whether their pubertal age onset was earlier than heterosexuals. In this analysis, they disregarded adolescent assessment of sexual orientation based on romantic attraction and relied on early adulthood sexual orientation identity. We recommend this course of action to researchers to discover the course of development for nonheterosexual young adults in Add Health.

Acknowledgments

This Invited Essay was requested by the Editor, who reviewed it and provided feedback to the authors. This research uses data from Add Health, a program project directed by Kathleen Mullan Harris and designed by J. Richard Udry, Peter S. Bearman, and Kathleen Mullan Harris at the University of North Carolina at Chapel Hill, and funded by grant P01-HD31921 from the Eunice Kennedy Shriver National Institute of Child Health and Human Development, with cooperative funding from 23 other federal agencies and foundations. Special acknowledgment is due Ronald R. Rindfuss and Barbara Entwisle for assistance in the original design. Information on how to obtain the Add Health data files is available on the Add Health website (http://www.cpc.unc.edu/addhealth). No direct support was received from grant P01-HD31921 for this analysis. The first author received financial support from the American Institute of Bisexuality and the second author received support from the Center for Family and Demographic Research, Bowling Green State University, which has core funding from the Eunice Kennedy Shriver National Institute of Child Health and Human Development (R24HD050959-09).

Copyright information

© Springer Science+Business Media New York 2013