PMCCPMCCPMCC

Search tips
Search criteria 

Advanced

 
Logo of nihpaAbout Author manuscriptsSubmit a manuscriptNIH Public Access; Author Manuscript; Accepted for publication in peer reviewed journal;
 
Pediatrics. Author manuscript; available in PMC Oct 27, 2009.
Published in final edited form as:
PMCID: PMC2768056
NIHMSID: NIHMS87172
Patient Teenagers? A Comparison of the Sexual Behavior of Virginity Pledgers and Matched Nonpledgers
Janet Elise Rosenbaum, PhD, AM
Janet Elise Rosenbaum, Health Policy PhD Program, Harvard University, Cambridge, Massachusetts; Department of Population, Family, and Reproductive Health, Johns Hopkins Bloomberg School of Public Health, Johns Hopkins University, Baltimore, Maryland;
Address correspondence to Janet Elise Rosenbaum, PhD, Department of Population, Family, and Reproductive Health, Johns Hopkins Bloomberg School of Public Health, 615 N Wolfe St., 4th Floor, Baltimore, MD 21205. E-mail: jerosenb/at/jhsph.edu
OBJECTIVE
The US government spends more than $200 million annually on abstinence-promotion programs, including virginity pledges. This study compares the sexual activity of adolescent virginity pledgers with matched nonpledgers by using more robust methods than past research.
SUBJECTS AND METHODS
The subjects for this study were National Longitudinal Study of Adolescent Health respondents, a nationally representative sample of middle and high school students who, when surveyed in 1995, had never had sex or taken a virginity pledge and who were >15 years of age (n = 3440). Adolescents who reported taking a virginity pledge on the 1996 survey (n = 289) were matched with nonpledgers (n = 645) by using exact and nearest-neighbor matching within propensity score calipers on factors including prepledge religiosity and attitudes toward sex and birth control. Pledgers and matched nonpledgers were compared 5 years after the pledge on self-reported sexual behaviors and positive test results for Chlamydia trachomatis, Neisseria gonorrhoeae, and Trichomonas vaginalis, and safe sex outside of marriage by use of birth control and condoms in the past year and at last sex.
RESULTS
Five years after the pledge, 82% of pledgers denied having ever pledged. Pledgers and matched nonpledgers did not differ in premarital sex, sexually transmitted diseases, and anal and oral sex variables. Pledgers had 0.1 fewer past-year partners but did not differ in lifetime sexual partners and age of first sex. Fewer pledgers than matched nonpledgers used birth control and condoms in the past year and birth control at last sex.
CONCLUSIONS
The sexual behavior of virginity pledgers does not differ from that of closely matched nonpledgers, and pledgers are less likely to protect themselves from pregnancy and disease before marriage. Virginity pledges may not affect sexual behavior but may decrease the likelihood of taking precautions during sex. Clinicians should provide birth control information to all adolescents, especially virginity pledgers.
Keywords: sexual abstinence, sexual behavior, sexual partners, contraception, adolescent, religion and sex, Christianity, nonparametric statistics, matched-pair analysis
What's Known on This Subject
Two studies have found, by using regression, that virginity pledges delay sex, but regression cannot correct for large preexisting differences between pledgers and nonpledgers.
What This Study Adds
We used a more robust method than regression to compare virginity pledgers with similar nonpledgers and found virtually no difference in sexual behavior or STDs and much less use of condoms.
Reducing early adolescent sexual initiation is an important public health objective. Early sexual initiation is associated with sexual risk-taking, pregnancy, and sexually transmitted diseases (STDs).1,2 Well-designed sex education programs that teach both abstinence and contraception can delay sexual initiation3,4 and prevent pregnancy, STDs, and risky sexual behavior.1,3-5 Abstinence-only sex education (AOSE) programs are defined by statute as having the “exclusive purpose [of] teaching the social, psychological, and health gains [of] abstaining from sexual activity.”6 No AOSE programs have been identified as changing adolescent sexual behavior in either the congressionally mandated randomized experiment7 or the systematic review of well-designed AOSE studies,8 but AOSE funding has increased dramatically, from $73 million in 2001 to $204 million in 2008.6,9
A sexual abstinence or “virginity” pledge is an oral or written promise to refrain from sexual activity, usually until marriage, administered after a multi- or single-session curriculum in religious youth groups, parochial and public schools, or large group events. The virginity pledge and 6-hour curriculum were created in 1993 by an evangelical Christian organization. The idea was subsequently spread by other Protestant and Catholic groups, which created pledges for their own AOSE programs for both religious and secular adolescents. By 1995, 13% of American adolescents reported having taken a virginity pledge.10 Virginity pledges are also now used to measure AOSE program effectiveness, which the US government considers successful if they produce many virginity pledgers, irrespective of participants' sexual behavior.6,11 This standard raises the question of whether virginity pledgers are less sexually active than comparable adolescents.
Studies using regression models have shown that virginity pledgers in a nationally representative sample were less likely to become sexually active than non-pledgers 1 and 5 years after the pledge,10,12,13 equally likely to have STDs 5 years after the pledge,10,13 and less likely to use contraception than nonpledgers.12
A California study showed greater sexual abstinence among adolescents who made a personal resolution not to have sex but not among formal virginity pledgers.14 These studies were critiqued for comparing pledgers with dissimilar nonpledgers.11
Regression models in past studies compared pledgers with the universe of nonpledgers despite dissimilarities that regression may be unable to correct.15-17 One year before pledging, pledgers are more religious, less sexually experienced, and hold more negative attitudes about sex and birth control than adolescents who do not go on to take a virginity pledge.10,12 Religious adolescents delay sexual initiation,18-21 so virginity pledgers' prepledge religiosity could induce abstinence without the pledge.
Given regression's recognized limitations,15-17 this article is distinctive in using matched sampling methods22-26 to compare the sexual and contraceptive behavior of virginity pledgers with similar nonpledgers in a national longitudinal study 5 years after a pledge is made.
Matched sampling is a nonparametric method for assessing program outcomes by comparing a program group with similar nonprogram respondents.22-24 We created a group of nonpledgers as similar as possible to pledgers on all prepledge factors that may influence sexual behavior, so outcome differences between pledgers and matched non-pledgers cannot be attributed to preexisting differences. Past studies compared self-selected virginity pledgers with the general population and attempted to adjust for the vast prepledge differences by using regression models.
Both matching and regression yield associative rather than causal inference, but matching creates more valid comparisons and results for 3 reasons. First, regression models rely on dubious parametric assumptions and cannot adjust, even on average, for large differences between program and nonprogram groups.15,16,24,25
Second, matching computes outcome differences only once, after verification that the matched nonprogram group is similar to the program group. This separation ensures that the model is selected independently of the study's results, in contrast to regression, with which it is impossible to verify model correctness without seeing the results. Third, matching allows adjustment for many more variables than does regression. In this study, I controlled for 112 variables, which would be problematic in a regression with 289 pledgers. For these reasons, matched sampling has been advocated for studies in medicine and public health27-29 and is used increasingly often in the medical literature.30-34
Data
Data are a subsample of the National Longitudinal Study of Adolescent Health (Add Health),* a nationally representative sample of grade 7 to 12 students interviewed in 3 waves (in 1995 [wave 1], 1996 [wave 2], and 2001 [wave 3]), as described elsewhere.35,36 The subsample comprises respondents who had not at wave 1 taken a virginity pledge or been sexually active, were >15 years of age, and participated in all 3 waves (n = 3440).
Respondents <15 years of age were not asked about sex and birth control attitudes, which likely influence both virginity pledge and sexual activity, so they were excluded from analysis.
Attrition in the subsample was 12% between waves 1 and 2 and 26% between waves 2 and 3, similar to that in the larger sample. Survey weights developed for the entire sample are inadequate for a constrained sub-sample and were not used.
Variables
Virginity-pledge status is the wave 2 answer to “Have you ever signed a pledge to abstain from sex until marriage?” Predictors of virginity pledge were measured at wave 1, 1 year before the pledge, and so cannot be attributed to the pledge as wave 2 factors could be. Factors for exact and nearest-neighbor caliper matching were selected from 128 potential predictors derived from past-pledge literature10,12-14 and the National Institute of Mental Health integrated health behavior model37: 16 composite variables, their 85 component survey items, and 27 other items. (Major categories and example items are listed in Table 1; the full list is available in the Appendix.)
TABLE 1
TABLE 1
Means for Pledgers Versus Nonpledgers, Before and After Matching
Outcome variables measure sexual behavior and STD diagnosis and prevention 5 years after the pledge, when the sample was at median 22 years of age (interquartile range: 20–23 years). STDs were diagnosed from urine-test results; all other outcomes were self-reported.
Respondents' use of STD- and pregnancy-prevention methods before marriage was measured by reported birth control and condom use in the past 12 months and at last sex and was coded as missing for married respondents.
Outcome differences were biased toward showing an effect of the pledge for 2 reasons: the pledge is an intermediate outcome of an unobserved abstinence intervention, which would be the treatment variable in an experiment; and pledgers may be less likely to report sexual activity than nonpledgers.38
Matched Sampling
Matched sampling attempts to create a group of nonpledgers with prepledge characteristics similar to pledgers, as would be true in a randomized experiment.17,22-24 Ideally, exact duplicates of every pledger could be found among the nonpledgers.31 Instead, matching creates a comparison group with a similar distribution of preprogram factors, which is thus the primary criterion for assessing match quality.23,24,26 There are not yet standardized guidelines for choosing a matching procedure,23,27 and matching methods yield similar results in simulation,23 but nearest-neighbor caliper matching is generally recommended.23 Exact matching can be combined with any method to make respondents identical on factors that might otherwise cause large differences, similar to blocking in randomized experiments.23,25 A matching method's appropriateness is gauged postfacto by the balance achieved, so any method and choice of matching factors that result in balanced groups is considered appropriate.23,24
This study used 2 types of matching: exact matching and 3:1 nearest-neighbor matching within propensity score calipers with replacement, using the R package MatchIt.25,39 The 2 exact matching factors were anticipating feeling guilty if they had sex and weekly attendance at church and/or religious youth group, conceptually distinct items with the largest mean prematching differences between pledgers and nonpledgers (Table 1). Propensity scores are the estimated probability of taking a virginity pledge calculated from a stepwise logistic regression.
Nearest-neighbor matching within propensity score calipers locates the 3 nonpledgers “closest” to each pledger, preferentially within calipers of 0.25 SDs in propensity score. The Mahalanobis metric measures the correlation-adjusted distance between respondents on the basis of respondents' values of continuous variables. The variables used in the distance measure are derived through trial and error by including and excluding variables until balance is achieved. Balance was achieved by using 4 composite variables in the metric: religious involvement, negative attitudes about birth control, parent religiosity, and pubertal development.
Match adequacy is determined by “balance,” the similarity of the covariate distributions of pledge and nonpledge groups. The t test is commonly used to assess balance at mean, but there is not yet consensus on the best way to assess balance across the entire distribution.25,26,30 This study assessed balance by using the t test and visual inspection of empirical quantile-quantile plots.26,30
Once balance was achieved, the outcomes of virginity pledgers and matched nonpledgers were compared with a t test. Cohen's effect size d, a measure independent of sample size, was calculated for significant differences: 0.2 is classified as small, 0.5 as medium.40
For illustrative comparison of differences between our restricted group and adolescents nationwide, survey-adjusted means of wave 3 outcomes were computed in Stata (Stata Corp, College Station, TX).
In ordinary regression, virginity pledgers would be compared with all nonpledgers, but these groups differed 1 year before taking the pledge. Comparing the 289 pledgers and 3151 nonpledgers at wave 1 before matching, pledgers were less sexually experienced and expected more negative and fewer positive psychosocial effects of sex and birth control use, with lower birth control efficacy and knowledge. Pledgers had greater levels of religious belief, involvement, Born Again affiliation, more religious parents, and fewer substance-using friends and were more likely to expect marriage before age 25. Pledgers also were disproportionately female, Asian, with foreign-born parents, and had lower Peabody vocabulary test scores (Table 1 and Appendix). Survey design parameters (region, cluster, and weight) were attempted as covariates in the propensity score model but were not significant.
The 3:1 matching with replacement matched 645 nonpledgers to the 289 pledgers. Matched pledgers and nonpledgers did not differ on average in propensity score, 16 composite variables and their 85 component items, and 27 other variables (Table 1 and Appendix). By simple chance, 5% of comparisons on average will be significant at the .05 level; 0 of the 128 comparisons are significant, so balance is better than expected by chance.
Turning to outcomes, 5 years after the pledge, 81.9% (confidence limits [CLs]: 76.2%, 87.6%) of virginity pledgers claimed to have never pledged. Virginity pledgers and matched nonpledgers did not differ in 12 of 14 sexual behaviors, 3 of 3 STD test results, and 4 of 4 marriage-related outcomes (Table 2). Pledgers reported an average of 1.09 past-year vaginal sex partners, 0.11 (CLs: 0.02, 0.19) fewer than nonpledgers, and 2.31% (CLs: 0.08%, 4.53%) fewer pledgers reported having been paid for sex than nonpledgers.
TABLE 2
TABLE 2
Sexual Behavior and Birth Control Use for Pledgers and Matched Nonpledgers, Wave 3
Unmarried pledgers were less likely to report using birth control and condoms in the last year, and birth control at last sex, but did not differ in reporting condom use at last sex or in condom breakage (Table 2).
The pledgers and matched nonpledgers together are a highly religious group of adolescents and would be expected to be more sexually conservative.18-21 Pledgers and matched nonpledgers together reported substantially more conservative sexual behavior at wave 3 than the general population of adolescents—with fewer reporting premarital vaginal sex, oral and anal sex, birth control and condom use, and multiple sex partners and more reporting being married—but did not differ in 2 of the 3 STD tests: fewer had positive test results for Neisseria gonorrhoeae but did not differ in the proportion testing positive for Chlamydia trachomatis or Trichomonas vaginalis compared with the general adolescent population in Add Health wave 3 (data not shown). Among wave 2-matched nonpledgers, 8.7% (CLs: 5.3%, 12.1%) reported a pledge at wave 3.
Pledgers were not less sexually active than matched nonpledgers despite prepledge similarities on 128 factors. Past findings that pledgers were less sexually active than the general population of nonpledgers may be attributable to regression models' failure to adjust for vast prepledge differences between the groups. Our refined sample (both pledgers and matched nonpledgers) is more religious and sexually conservative than the general population of adolescents and would be predicted to delay sex without virginity pledges.18-21
Despite having had similar birth control attitudes 1 year before pledging, virginity pledgers were substantially less likely than matched nonpledgers to protect themselves against STDs and pregnancy, consistent with earlier studies.10,12
Virginity pledgers may be less likely to use condoms and contraception because many abstinence programs cause participants to develop negative attitudes about their effectiveness.7,41
More than 90% of abstinence funding does not require that curricula be scientifically accurate,6,9 and a 2004 review found incorrect information in 11 of 13 federally funded abstinence programs, primarily about birth control and condom effectiveness.42
Most virginity pledgers reported having had premarital vaginal and oral sex but did not seem to substitute oral and anal sex for vaginal sex, contrary to earlier studies.10 Virginity-pledge programs do not prepare pledgers to protect their health if they have sex, although most pledgers do have sex. Pledge programs have guidance for pledgers who initiate sex, such as the True Love Waits publication When True Love Doesn't Wait,43 the recommendations of which include a medical examination and a second, mentored pledge.
Virginity pledgers have 0.1 fewer past-year sexual partners on average, but this modest difference is unlikely to affect STD risk, because pledgers do not differ in the average number of lifetime partners (~3 each) or age of sexual initiation (age 21) or in empirical STD prevalence.
Few virginity pledgers continue to identify with their pledges 5 years after pledging, with >80% claiming to have never pledged, consistent with an earlier finding that half of pledgers disaffiliated within 1 year.38 This high rate of disaffiliation may imply that nearly all virginity pledgers view pledges as nonbinding.
Limitations
Matching adjusts only for observed characteristics, but the finding of no difference is robust to matching adequacy. Even if unobserved differences remained after matching, the data could falsely indicate no difference between groups only if pledgers were less abstinent than nonpledgers. Differences between groups may be attributable to an unobserved characteristic, but large differences such as a 10 percentage-point difference in past-year condom use require finding unobserved characteristics with more effect than the 128 factors already matched on, which is unlikely.44
Outcome differences are biased toward showing a pledge effect, because the pledge is an intermediate variable to an unobserved treatment variable: abstinence education program participation, unmeasured in the Add Health survey. Approximately 5% of the 32 outcomes compared may be statistically significant by chance because of multiple comparisons. These biases are unlikely to cause a full 10 percentage-point difference at all levels of condom use.
Sexual behavior reports are likely biased toward showing a pledge effect because virginity pledgers may under-report sex38; failure to observe a difference in sexual behavior reinforces the likelihood of no true difference.
Pledges were taken in 1996, but the prevalent pledge text and curriculum have not changed substantially since then, according to virginity pledge co-creator (Rev. James Hester, personal communication, September 20, 2007). Pledge programs differ in their educational programs, continued contact with pledgers, and possible effectiveness, but this study cannot differentiate among programs and computes an average difference over all programs.
Measurement is of self-reported virginity pledges, but within 1 year, half of the virginity pledgers denied having pledged.38 Adolescents who pledged and ended identification with the pledge before wave 2 were counted as nonpledgers, and 8.7% of wave 2 nonpledgers reported a pledge at wave 3; these bias results to show no pledge effect.
Adolescents were ≥15 years of age at wave 1 because of unavailability of sex and contraceptive attitude data for younger adolescents. Younger virginity pledgers may be more likely to delay sex over a period of 1 year, as a previous study found,12 but as was true for older adolescents, part of the delay is likely attributable to pledgers' prepledge attitudes, not the pledge.
Premarital sex, condom, and birth control use cannot be detected for married respondents and were not imputed. Anal and oral sex prevalence come from respondents' descriptions of each of their past relationships and, thus, are likely to be underestimates: 75% of respondents reported having had vaginal sex, but only 63.8% of respondents reported vaginal sex in describing past relationships.
This article maximizes internal validity, with sacrifice to external: the restricted subsample is not nationally representative.
Policy Implications
The results suggest that the virginity pledge does not change sexual behavior. One cannot make causal inferences given the pledge's voluntary nature, but if the pledge decreased sexual activity, we would expect to observe a difference between virginity pledgers and comparable nonpledgers; indeed, this estimate is biased in favor of showing a pledge effect.
Given this evidence that pledgers are less likely than comparable nonpledgers to use condoms and birth control, and previous evidence that AOSE programs do not affect sexual behavior,7,8 federal AOSE funds should be shifted to evidence-based sex education programs that teach birth control and have been demonstrated to delay sexual initiation3,4 and increase safer sex practices.1,3-5
Virginity pledges are not a marker for less sexual activity and should not be used as a measure of abstinence sex education program effectiveness.
CONCLUSIONS
Adolescents who take virginity pledges are not less sexually active than closely matched adolescents who do not take pledges, but they are less likely to use birth control and condoms. Clinicians should provide birth control information to all adolescents, especially AOSE participants.
TABLE 4
TABLE 4
Factors From the National Institute of Mental Health Integrated Health Behavior Model: Intention to Abstain
TABLE 8
TABLE 8
School Integration, Positive Expectancies, Self-efficacy
TABLE 9
TABLE 9
Parent Religiosity, Respondent Religiosity, Romantic Experience, Friend Birth Control Knowledge, and Friend Risk Behavior
ACKNOWLEDGMENTS
This work was supported by the Milton Fund of Harvard Medical School (Michael Ganz, principal investigator), the Harvard Graduate School of Arts and Sciences, the Harvard PhD Program in Health Policy, the Department of Society, Human Development, and Health at the Harvard School of Public Health, and the Institute for Health Research and Policy at the University of Illinois at Chicago.
I am grateful for the guidance of my dissertation committee: Joanne Cox, Don Rubin, Kimberly Thompson, and Alan Zaslavsky. I thank the following individuals for helpful discussions: Bob Blum, Barbara Devaney, Michael Ganz, Mark Goldstein, Jim Greiner, David Hemenway, Jim Hester, Charles Horn, Olivia Lau, Rebekah Maggor, Richard Ross, Joe Schafer, Kathy Swartz, Chris Trenholm, Chris Winship, Laurie Zabin, and anonymous reviewers.
Abbreviations
STDsexually transmitted disease
AOSEabstinence-only sex education
CLconfidence limit

APPENDIX
Tables Tables33 through through55 show factors considered as potential predictors of taking a virginity pledge, categorized according to their origin. Potential predictors of taking a virginity pledge were derived from the past virginity-pledge literature and the National Institute of Mental Health integrated model for health behavior. Potential predictors are listed under model components as an aid to the reader, although many other categorizations are possible. Covariates were added separately to the logistic regression model in a “flat” manner, so their categorization did not influence the logistic regression model for predicting probability of pledging. All variables were included in the propensity score model separately, rather than in index format. Variables were combined into indices so that they are continuous for use in the nearest-neighbor matching. All binary variables were defined as 1 if endorsed and 0 otherwise.
TABLE 3
TABLE 3
Factors From the Literature Considered As Potential Predictors of Virginity Pledge
TABLE 5
TABLE 5
Factors From the National Institute of Mental Health Integrated Health Behavior Model: Knowledge/Skills, Absence Environmental Constraints, Salience, Habit
Table 6 shows the logistic regression model results predicting taking a wave 2 virginity pledge from wave 1 characteristics. The logistic regression model was determined from stepwise logistic regression in Stata by using the factors listed in Tables Tables33 through through55.
TABLE 6
TABLE 6
Logistic Regression Results: Prepledge Characteristics (Wave 1) Associated With Taking a Virginity Pledge (Wave 2)
Tables Tables77 through through1010 show the means before and after matching for each of the variables listed in Tables Tables33 through through5.5. These tables were extracted into Table 1.
TABLE 7
TABLE 7
Demographics and Family Background: Before and After Matching
TABLE 10
TABLE 10
Learned Health Behavior in School: Attitudes Toward Sex and Birth Control
Variable Selection
The following details how variables were selected by using the past pledge literature11-14 and the National Institute of Mental Health integrated health behavior model.33
The past pledge literature was used for identifying variables associated with pledging. I attempted to use all available variables in every article that examined predictors of pledging and all significant variables.
Variable Coding and Missing Data
All items were coded for endorsement or nonendorsement; skipping a question counted as nonendorsement. Peabody vocabulary scores and parent-reported household incomes were median-imputed, and an indicator for missingness was created. Height and weight were imputed by using regression on age, gender, interaction, and each other and median-imputed for the remaining cases. Median imputation artificially decreases SE and may cause a nonsignificant factor to be included in the matching model, but that does not compromise match quality. Sex and birth control outcomes for respondents who never had sex were coded as missing. The number of lifetime and past-year sex partners was truncated at the 90th percentile for each gender: females at 8 lifetime and 2 past-year partners and males at 10 and 3, respectively. Past-year and lifetime partners and age of first sex at wave 3 were each skipped by <1% of respondents; number of times the respondent had sex in the past year was skipped by 7% of respondents. These continuous outcomes were regression-imputed by using wave 1 age, gender, race/ethnicity, church attendance, and parent education and income as predictors. Because pregnancy and STD prevention are generally unimportant for married respondents, these questions were coded as missing for respondents unmarried at wave 3.
Footnotes
The author has indicated she has no financial relationships relevant to this article to disclose.
*Add Health is a program project designed by J. Richard Udry, Peter S. Bearman, and Kathleen Mullan Harris, and funded by National Institute of Child Health and Human Development grant P01-HD31921, with cooperative funding from 17 other agencies. Special acknowledgment is due to Ronald R. Rindfuss and Barbara Entwisle for assistance in the original design. Persons interested in obtaining data files from Add Health should contact Add Health, Carolina Population Center, 123 W Franklin St, Chapel Hill, NC 27516-2524 (www.cpc.unc.edu/addhealth/contract.html).
This research was approved by the Harvard University Human Subjects Board. Dr Rosenbaum conceived the project, analyzed the data, interpreted the results, and wrote the article.
This article was presented in part at an Association for Public Policy Analysis and Management meeting (poster), November 7, 2008, Los Angeles, CA; an American Public Health Association meeting, October 28, 2008, San Diego, CA; the R User Conference, August 12, 2008, Dortmund, Germany; Joint Statistical meetings, August 3, 2008, Denver, CO; George Washington University, June 25, 2008, Washington, DC; the International Conference on Health Policy Research, January 17, 2008, Philadelphia, PA; and a Johns Hopkins STD Center research seminar, December 19, 2007, Baltimore, MD.
1. Kirby D. Emerging Answers 2007: Research Findings on Programs to Reduce Teen Pregnancy and Sexually Transmitted Diseases. National Campaign to Prevent Teen Pregnancy; Washington, DC: 2007.
2. Sandfort TG, Orr M, Hirsch JS, Santelli J. Long-term health correlates of timing of sexual debut: results from a national US study. Am J Public Health. 2008;98(1):155–161. [PubMed]
3. Santelli J, Ott MA, Lyon M, Rogers J, Summers D, Schleifer R. Abstinence and abstinence-only education: a review of US policies and programs. J Adolesc Health. 2006;38(1):72–81. [PubMed]
4. Kirby DB, Laris BA, Rolleri LA. Sex and HIV education programs: their impact on sexual behaviors of young people throughout the world. J Adolesc Health. 2007;40(3):206–217. [PubMed]
5. Underhill K, Operario D, Montgomery P. Systematic review of abstinence-plus HIV prevention programs in high-income countries. PLoS Med. 2007;4(9):e275. Available at: http://medicine.plosjournals.org/perlserv/?request=get-document&doi=10.1371/journal.pmed.0040275&ct=1. Accessed January 14, 2008. [PMC free article] [PubMed]
6. Government Accountability Office Abstinence Education: Efforts to Assess the Accuracy and Effectiveness of Federally-Funded Programs, October 2006. GAO-07-87. Available at: www.gao.gov/cgibin/getrpt?GAO-07-87. Accessed October 15, 2007.
7. Trenholm C, Devaney B, Forston K, Quay L, Wheeler J, Clark M. Impacts of four Title V, Section 510 abstinence education programs. Mathematica Policy Res. Apr, 2007. Available at: www.mathematica-mpr.com/publications/PDFs/impactabstinence.pdf. Accessed January 14, 2008.
8. Underhill K, Operario D, Montgomery P. Sexual abstinence only programmes to prevent HIV infection in high income countries: systematic review. BMJ. 2007;335(7613):248. [PMC free article] [PubMed]
9. Department of Health and Human Services HHS Budget in Brief 2008. Available at: www.hhs.gov/budget/08budget/2008BudgetInBrief.pdf. Accessed June 23, 2008.
10. Bearman PS, Brückner H. After the promise: the STD consequences of adolescent virginity pledges. J Adolesc Health. 2005;36(4):271–278. [PubMed]
11. Constantine NA, Braverman MT. Appraising evidence on program effectiveness. In: Braverman MT, Constantine NA, Slater JK, editors. Foundations and Evaluation: Contexts and Practices for Effective Philanthropy. Jossey-Bass; San Francisco, CA: 2004. pp. 236–258.
12. Bearman PS, Brückner H. Promising the future: virginity pledges and first intercourse. Am J Sociol. 2001;106(4):859–912.
13. Ford CA, Pence BW, Miller WC, et al. Predicting adolescents' longitudinal risk for sexually transmitted infection: results from the National Longitudinal Study of Adolescent Health. Arch Pediatr Adolesc Med. 2005;159(7):657–664. [PubMed]
14. Bersamin MM, Walker S, Waiters ED, Fisher DA, Grube JW. Promising to wait: virginity pledge and adolescent sexual behavior. J Adolesc Health. 2005;36(5):428–436. [PMC free article] [PubMed]
15. Lalonde R. Evaluating the econometric evaluations of training programs. Am Econ Rev. 1986;76(4):605–620.
16. Dehejia HR, Wahba S. Causal effects in nonexperimental studies: reevaluating the evaluation of training programs. J Am Stat Assoc. 1999;94(448):1053–1062.
17. Rosenbaum PR. Matching in observational studies. In: Gelman A, Meng XL, editors. Applied Bayesian Modeling and Causal Inference From Incomplete-Data Perspectives. John Wiley and Sons; Chicester West Sussex, England: 2004. pp. 5–24.
18. Miller L, Gur M. Religiousness and sexual responsibility in adolescent girls. J Adolesc Health. 2002;31(5):401–406. [PubMed]
19. Nonnemaker JM, McNeely CA, Blum RW. Public and private domains of religiosity and adolescent health risk behaviors: evidence from the National Longitudinal Study of Adolescent Health. Soc Sci Med. 2003;57(11):2049–2054. [PubMed]
20. Jones RK, Darroch JE, Singh S. Religious differentials in the sexual and reproductive behaviors of young women in the United States. J Adolesc Health. 2005;36(4):279–288. [PubMed]
21. Lain MT, Wang J, Barry M. A longitudinal study of adolescent transition from virgin to nonvirgin status. J Adolesc Health. 2008;42(3):228–236. [PubMed]
22. Rubin DB. Matched Sampling for Causal Effects. Cambridge University Press; New York, NY: 2006.
23. Morgan SL, Winship C. Counterfactuals and Causal Inference: Methods and Principles for Social Research. Cambridge University Press; New York, NY: 2007.
24. Gelman A, Hill J. Data Analysis Using Regression and Multilevel/Hierarchical Models. Cambridge University Press; New York, NY: 2007.
25. Ho D, Imai K, King G, Stuart E. Matching as nonparametric preprocessing for reducing model dependence in parametric causal inference. Polit Anal. 2007;15(3):199–126. Available at: http://gking.harvard.edu/utiles/abs/maTchp-abs.shtml. Accessed May 4, 2008.
26. Imai K, King G, Stuart EA. Misunderstandings among experimentalists and observationalists about causal inference. J R Stat Soc Ser A Stat Soc. 2008;171(2):481–502. Available at: http://imai.princeton.edu/research/balance.html. Accessed May 7, 2008.
27. Rubin DB. Estimating causal effects from large data sets using propensity scores. Ann Intern Med. 1997;127(8 pt 2):757–763. [PubMed]
28. Little RJ, Rubin DB. Causal effects in clinical and epidemiological studies via potential outcomes: concepts and analytical approaches. Annu Rev Public Health. 2000;21:121–145. [PubMed]
29. Rubin DB. The design versus the analysis of observational studies for causal effects: parallels with the design of randomized trials. Stat Med. 2007;26(1):20–36. [PubMed]
30. Austin PC. A critical appraisal of propensity-score matching in the medical literature between 1996 and 2003. Stat Med. 2008;27(12):2037–2049. [PubMed]
31. Stuart EA. Developing practical recommendations for the use of propensity scores: discussion of a critical appraisal of propensity-score matching in the medical literature between 1996 and 2003. Stat Med. 2008;27(12):2062–2065. [PubMed]
32. Hansen BB. The essential role of balance tests in propensity-matched observational studies: comments on a critical appraisal of propensity-score matching in the medical literature between 1996 and 2003. Stat Med. 2008;27(12):2050–2054. [PubMed]
33. Hill J. Discussion of research using propensity score matching: comments on a critical appraisal of propensity-score matching in the medical literature between 1996 and 2003. Stat Med. 2008;27(12):2055–2061. [PubMed]
34. Austin PC, Grootendorst P, Anderson GM. A comparison of the ability of different propensity score models to balance measured variables between treated and untreated subjects: a Monte Carlo study. Stat Med. 2007;26(4):734–753. [PubMed]
35. Udry JR. The National Longitudinal Study of Adolescent Health (Add Health), Waves I & II, 1994–1996; Wave III, 2001–2002 [machine-readable data file and documentation] Carolina Population Center, University of North Carolina; Chapel Hill, NC: 2003.
36. Tourangeau R, Shin HC. National Longitudinal Study of Adolescent Health: Grand Sample Weight. Carolina Population Center, University of North Carolina; Chapel Hill, NC: 1999.
37. Jaccard J, Dodge T, Dittus P. Parent-adolescent communication about sex and birth control: a conceptual framework. New Dir Child Adolesc Dev. 2002;97(97):9–41. [PubMed]
38. Rosenbaum JE. Reborn a virgin: adolescents' retracting of virginity pledges and sexual histories. Am J Public Health. 2006;96(6):1098–1103. [PubMed]
39. Ho D, Imai K, King G, Stuart E. MatchIt: Nonparametric Preprocessing for Parametric Causal Inference. Version 2.3-1. Available at: http://gking.harvard.edu/matchit. Accessed May 13, 2008.
40. Cohen J. Statistical Power Analysis for the Behavioral Sciences. 2nd ed. Lawrence Erlbaum Associates; Hillsdale, NJ: 1988.
41. Borawski EA, Colabianchi N, Levers-Landis CE, et al. First year outcomes of Healthy Teens Building Healthy Schools: an abstinence-until-marriage vs comprehensive sex education intervention study [abstract]; Presented at: the American Public Health Association 133rd annual meeting; Philadelphia, PA. December 10-14, 2005; Abstract 108124.
42. United States House of Representatives Committee on government reform, minority staff special investigations division. The Content of Federally-Funded Abstinence-Only Education Programs. Prepared for Rep. Henry A. Waxman, December 2004.
43. Rankin DT. When True Love Does'nt Wait. LifeWay Christian Resources; Nashville, TN: 1998.
44. Rosenbaum PR. Observational Studies. 2nd ed. Springer; New York, NY: 2002.