Estimating the longitudinal association between adolescent sexual behavior and exposure to sexual media content (2009)

Study Design and Participants

Data collection took place via a Web-based survey fielded during the spring and summer of 2005, 2006, and 2007. Adolescent respondents were recruited through print and radio advertisements, direct mail, and word of mouth to complete the survey. Recruitment was best achieved for Black respondents (49%) by Metro ads (Metro is a free newspaper distributed via street corner bins and on the public transportation system in Philadelphia) followed by word of mouth (14%) or an unknown method (14%). White and Hispanic respondents showed a more equal mix of methods. The three best methods for White respondents were Metro ads (27%), through respondents earlier recruited (23%), and via direct mail (14%). The three best methods for Hispanic respondents were Metro ads (28%), through respondents earlier recruited (23%), and word of mouth (13%).

Respondent eligibility criteria included age at the time of the initial survey (14, 15, or 16) and race/ethnicity (White, African-American, or Hispanic). The sampling strategy was quota-driven with a desire for roughly equal sample sizes in all Race*Age*Gender cells (a 3*3*2 design). In practice, adolescent Hispanic respondents in the Philadelphia metropolitan area were extremely difficult to locate and recruit, so their cell frequencies are low. The survey was launched in April 2005 following a test of the technology and a pre-test of the survey instrument. Incompletion rates (e.g., the number not completing the first survey divided by the number successfully consented) were similar for Black and Hispanic respondents (17% and 19% respectively) and lower for White respondents (6%). There was no difference in incompletion rates by gender (males = 14%, females = 13%).

The survey was accessible from any computer with internet access. Participants were given the option of taking the survey at the University or an off-site location (e.g. home, school, or community library). Respondents were assigned a password to access the survey, as well as an identification number and personal password to ensure confidentiality and privacy protection. Respondents were compensated $25 dollars upon completion of the survey at each wave, and on average, took one hour to complete the survey. Those respondents who completed all 3 waves of the survey received a bonus of $25. After submitting respondent assent/parental consent forms, 547 adolescents ages 14 to 16 completed the survey at Wave 1 (in 2005). There are a small number of missing values although retention rates over the three waves of data collection were high (87% of the initial sample were successfully recontacted in all waves and 94% of the initial sample participated in at least 2 of 3 waves) and the data set used here is limited to 506 respondents who are present in the data set for at least 2 of the 3 waves of data collection. The respondents are 62% female, 42% African-American, 42% White, 13% Hispanic, and 3% “Other.” For White respondents, the sample sizes in year 1 by age (14, 15, and 16) were 67, 73, and 73 respectively, for Black respondents were 74, 76, and 73 respectively.

Dependent Variable: Sexual Behavior Index Score

The survey collected data for lifetime, more than a year ago, and during the past 12 months on the following sexual behaviors: deep kissing (item: Have you ever participated in deep kissing (some people call this “French kissing”)?), touching the breasts of a female partner (item: If you ever have had a female partner, did you touch her breasts?), the respondent having their breasts touched (item: Have you ever had your breasts touched by a partner?), genital touching of the respondent by a partner (item: Has a partner ever touched your private parts?), the respondent giving oral sex (item: Have you ever put your mouth on a partner’s private parts (some people cal this “oral sex”)?), the respondent receiving oral sex (item: Has a partner ever put their mouth on your private parts (some people call this “oral sex”)?), the respondent receiving anal sex (item: If you ever had a male partner, has he ever put his penis in your anus (some people call this “anal sex”)?), the respondent giving anal sex (asked of males only, item: Have you ever put your penis in your partner’s anus (some people call this “anal sex”)?), and having vaginal sexual intercourse (item: Have you ever had sexual intercourse (i.e., a penis in the vagina) with a partner of the opposite sex?).

Because the age range is limited, we focus on the lifetime items because many of the behaviors are rare or zero using shorter recall periods. We limit the analyses to heterosexual behaviors, so the breasts touched and the receiving anal sex variables are used only with females, and the touching breasts and giving anal sex variables are used only with males. We also drop from the analysis 6 males who have received anal sex because their inclusion reduces the hierarchical nature of the index for males. These respondents did not report any lifetime occurrence of vaginal sex so our heterosexual behavioral index is probably inappropriate for them.

We used Mokken scaling to assess the scalability of the dichotomous sexual behavior items. Mokken scaling is based on difficulty ordering, such that all items after the initial failure are also failed and all items before the initial failure are passed (Ringdal, Ringdal, Kaasa, Bjordal, Wisløff, Sundstrøm & Hjermstad, 1999). If the items scale using this definition, the items are considered difficulty ordered, and the researcher knows precisely what a “2” (for example) on the index score means. In this instance, a respondent with a “2” performed the first two sexual behavior items and did not perform the last 5. This is the interpretive advantages of difficulty ordering: the value of the summative index indicates which items were passed and which were failed. Scaling sexual behaviors in this way provides researchers with an index reflecting a “sexual behavior hierarchy.”

Item sets are evaluated for difficulty-ordered unidimensionality using Loevinger’s H coefficient (Ringdal et al. 1999); a value of .5 or more indicates a strong scale (Mokken, 1971, p. 185). For each year, the items scaled well: H for males was 0.75 in year 1, 0.70 in year 2, and 0.77 in year 3; H for years 1 to 3 for females was 0.83, 0.84, and 0.83 respectively. Average sexual behavior index scores by study wave were 2.71 (SD = 2.23), 3.62 (SD = 2.26), and 4.46 (SD = 2.17) for years 1, 2, and 3 respectively. The ordering of the behaviors across genders was: deep kissing, touching breasts/breasts touched, genital touching, receiving oral sex, vaginal sex, giving oral sex, and receiving/giving anal sex. However, for males in year 2, the order of receiving oral sex (45%) and reporting vaginal sex (44%) is reversed (by 1%) compared with Year 1. In year 3 the order for males is identical to year 1. For females, the order of the behaviors is consistent across all three data collection years. More detail on difficulty-ordering as applied to these data can be found in Hennessy, Bleakley, Fishbein & Jordan (2008).

Dependent Variable: Exposure to Media Sexual Content

Our measure of exposure to media sexual content was calculated based on 2 types of variables: the respondents’ self reported exposure to selected media titles in 4 media (television, music, magazines, and videogames) and the respondents’ rating of sexual content in each of the media titles. The lists were constructed to reflect popular titles for teenagers and/or the general public at the time of the baseline survey and were updated for years 2 and 3. Popular titles were provided by website rankings (including: www.top5s.com/tvweek; www.boxofficemojo.com; www.imdb.com/boxoffice/rentals; www.billboard.com; www.gamerankings.com) and from an audience research company (TRU data) as well as pilot surveys we conducted in the year prior to the launching of the survey. The titles were designed to provide a sense of the depth and breadth of media use, although we recognized that they could not capture all that teens viewed, played, or read. In year 1 of the study, the survey included lists of 30 television programs, 30 music artists, 20 magazine titles, and 15 videogames. In year 2 of the study the list included 75 television titles, 50 music artists, 30 magazines, 40 movies, and 40 videogames, and in year 3 of the study the list included 74 television shows, 39 music artists, 32 magazines, 43 movies, and 45 videogames. For this analysis, however, movies are not included in the all media sexual content exposure measures to maintain comparability across time.

Self-reports, rather than the results of content analysis, were used because only self-report measures were collected during all three years of the survey. However, the correlations between exposure measures based on respondents own sex content ratings (as used here) and those based on the content analysis sex content ratings (for years 1 and 2, there was no content analysis of year 3 titles) were positively correlated (r = .75 in year 1 and r = .77 in year 2). In addition, the correlations between the exposure measure based upon self-reported sexual content and the sexual behavior index were very similar to the correlations between the exposure measure based upon the content analytic ratings of sex content and the sexual behavior index: The year 1 correlation between the sexual behavior index and the respondent-based exposure measure was .20 (p < .01) while the year 1 correlation between the sexual behavior index and the content-analysis-based exposure measure was .23 (p < .01).

Using an ordinal measure of exposure on a 4-point scale (never, rarely, sometimes, often) respondents indicated how frequently within the last 12 months they watched each show, listened to each artist, read each magazine, and played each videogame. Next, the respondents were asked to assess the sexual content of those same titles based on the following definition of sexual content: “In this survey, sexual content is defined as talking about or showing: hooking-up/making out; sexy clothes; nudity; sex (oral, anal, or vaginal); safe sex (condoms, birth control, etc.); sex crimes (rape); homosexuality (gay or lesbian); or anything else sex related.” This definition appeared in each media section in the survey immediately prior to the set of questions for which respondents were asked to rate the sex content of the media titles. In answer to the question, “How would you rate the sexual content of the following…,” adolescents rated the sexual content of the all media titles on a 4-point scale with the following responses: “no sexual content,” “a little sexual content,” “some sexual content,” and “a lot of sexual content.” Another response option, “I don’t know/I don’t watch this show,” was also included since respondents were asked to rate the sexual content of each title even if they indicated earlier that they had never been exposed to that particular media title. However, only the sexual content of the media titles to which they were exposed are included in our measure of exposure to sexual media content.

To calculate the sexual content exposure measure, the cross product of the exposure measure and the sexual content rating for each title was summed within each type of media, resulting in sex content exposure measures specific to television, music, magazines, and videogames. The total media sexual content exposure measure was created by summing the 4 media specific measures. For the exposure measure in the current analysis, the square root transformation was imposed on the total measure to better approximate a normal distribution and then this variable was transformed into Z scores. The latter adjustment is necessary because otherwise there could be larger values in later years purely due to the fact that more media titles were evaluated in years 2 and 3. Thus, for all three waves of the study, average exposure scores were 0 with a standard deviation of 1 (note that this transformation does not imply that exposure scores by age group all have identical means, see Figure 1 below). Pearson correlations between total sexual content exposure in year 1 and year 2 was r = 0.61 ( p < .05) and for year 2 and year 3 was r = .68 ( p < .05). Additional validity information of for the exposure to sexual content in the media measure is available elsewhere (Bleakley, Fishbein, Hennessy, Jordan, Chernin & Stevens, 2008).

  

Figure 1

Sexual Behavior Index and Exposure Score

What are Growth Curves?

Growth curve analysis is a statistical method for measuring change over time in an outcome variable (Curran & Hussong, 2002; Karney & Bradbury, 1995). It assumes that change is a continuous process, so estimating the slope of change over time in the dependent variable is the primary research question (Curran & Muthen, 1999). Non-time dependent variables (e.g., gender, experimental status, and race/ethnicity) can be included as predictors for statistical adjustment purposes or to investigate interactions between change over time and these fixed characteristics.

The Unconditional Equation

The Conditional Equation

One important implication of the growth curve approach is that because the intercept (η_i0) and slope (η_i1) parameters of the equation vary across individuals (note the respondent specific subscripts on the intercept and slope parameters in equation (1) above, something that never occurs in “ordinary” regression), they can be treated as dependent variables in auxiliary equations that predict the initial value and slope of the outcome. Known as a “conditional growth model,” the auxiliary equations predict the parameters (e.g., the intercept and slope) of the individual equation (Bollen & Curran, 2006, p. 9). Here we use gender and ethnicity as predictors to address research question number 5.

For all analyses we estimate the unconditional and conditional models simultaneously — a “parallel process” growth model (Cheong, MacKinnon & Khoo, 2003). This type of growth model allows for the estimation of the correlation between changes in sexual behavior and changes in exposure to sexual media content, as well as the correlations between the parameters of each equation.

Defining the Time Variable

Although the ASAMS project collected data for three years, this longitudinal structure (i.e., wave of the study) is not appropriate because the planned variability in respondent’s age at the beginning of the study confounds wave of study with age of respondent. That is, in each of the three waves of the study, respondents of three different ages are arbitrarily combined in a way that is not meaningful because “wave of the study” is a logistical feature of the data collection process: the age of the respondent is the primary developmental predictor (Bollen & Curran, 2006, pp. 79-81; Singer & Willett, 2003, p. 139). Confounding study wave and age of the respondent could easily have negative consequences since sexual behavior, at least, is positively related to age. Thus, to unconfound the respondent’s age with the study wave, we reorganize the data into an “accelerated cohort” design (Duncan, Duncan, Strycker, Li, & Alpert, 1999, Chapter 6; Raudenbush & Chan, 1992) so that age of the respondent is the longitudinal variable of interest. The result is five years of data ranging from 14 years olds in the first wave of the study to 18 year olds in the last wave, even though no respondent has more than three observations in the data set as a whole.

Growth Curve Results: Estimating the Best Fitting Time Metric

Analyses of the unconditional models (not shown) with a time metric free to vary (Biesanz, Deeb-Sossa, Papadakis, Bollen & Curran, 2004) shows that a linear model for age is an excellent fitting time metric for the sexual index outcome. Thus, for this equation the time metric is defined as Age minus 14 or 0 through 4 (e.g., 14-14 = 0; 15-14 =1; 16-14 =2, etc.). This metric makes the intercept term the predicted sex index score for fourteen year olds. Because the metric is linear, the change from 14 to 16 is twice as large as the change from 14 to 15, and the change from 14 to 18 is four times as large. However, the best fitting time metric for exposure was non-linear and a good fitting metric suggested by analyses allowing the time metric to vary is 0, 1, 1.5, 2, 2.25. Here, the change from 14 to 16 is only 1.5 times as large as the change from 14 to 15, and the change from 14 to 18 is only 2.25 times as large as the change from 14 to 15. In this non-linear case, if the slope of exposure over time is positive, this time metric produces a positive slope that flattens with increasing age, but if the slope of exposure over time is negative, a negative slope that flattens with increasing age is estimated.

Unconditional Growth Curve Results

Table 1 presents the results of the unconditional growth curve analysis. The fit of the model is good. The sexual index results show a predicted equation of 1.82 + .89(Time). The 1.82 is the predicted value of the sex index for 14 year olds and the slope of .89 indicates an increase of almost one sexual index unit per age increment for the sample as a whole. The negative correlation between the intercept and the slope indicates that the higher the initial value of the sex index, the lower the slope of change, that is, the slower the increase in sexual behavior by age. This is a reasonable result given the ceiling effects of an index that goes from 0 to 7. The intercept and slope have significant variation, so there are between subject differences in these parameters that might be explained by respondent characteristics

  

Table 1

Results for Unconditional Parallel Process Growth Mode of Sex Score Index and Exposure to Sexual Media Content (N = 506)

The unconditional equation for exposure to sexual content is -.041 + .025 (Time) indicating lower than average exposure for 14 year olds compared with older respondents and a positive increase in exposure to sexual content over time, although neither the intercept nor the slope is significantly different from zero. However, both parameters have significant variation which suggests that the respondents are variable over time and the unconditional average results may not necessarily be representative of certain subgroups. The negative correlation between the intercept and slope for exposure to sexual media content indicates that higher the initial value of exposure, the slower the increase in exposure to sexual content over time.

Research questions 2, 3, and 4 are answered by the intercept/slope correlations across the two equations. The correlation for initial value of exposure predicting the slope of sexual behavior is -.14 (p > .05) and the correlation between the initial value of sexual behavior with the slope of exposure to sexual media content is -.21 ( p < .05). For the sample as a whole then, although the initial value of exposure does not predict changes in sexual behavior, the initial value of sexual behavior does predict change in exposure, with higher initial values of sexual behavior being associated with slower increases in exposure to sexual content over time. Finally, the correlation between the two slope values is .09, which is positive but not discernable from zero. For the sample as a whole, changes in exposure to sex content and changes in sexual behavior are essentially unrelated. However, the conditional analysis, presented below, presents a very different picture of the complex relationship between sexual behavior and exposure to sexual media content.

This publication was made possible by Grant Number 5R01HD044136 from the National Institute of Child Health and Human Development (NICHD). Its contents are solely the responsibility of the authors and do not necessarily represent the official views of the NICHD.