Sampling television programs for content analysis of sex on TV: how many episodes are enough?

Researchers regularly conduct studies of sexual content on television using content analysis methods. Content analyses of television programs typically are conducted to examine what sexual messages people are exposed to (e.g., Kunkel, Eyal, Finnerty, Biely, & Donnerstein, 2005; Collins et al., 2004; Pardun, L'Engle, & Brown, 2005) and to explore cultural norms about sex (e.g., Ward, 1995). These studies make important contributions to our understanding of the kinds of sexual messages to which adolescents may be exposed by systematically analyzing a dominant medium for youth: television.

Analyses of television's messages about sexuality often are driven by the assumption that television acts as a kind of "super peer" as youth acquire knowledge, information, and beliefs about who is sexual, when it is appropriate to behave sexually, and the risks and responsibilities of becoming sexually active. These assumptions are likely justified. In 2000, the Kaiser Family Foundation reported that, among 1,510 12- to 18-year-olds, 23% reported television and movies as providing "a lot" of information about pregnancy and birth control (in Strasburger & Wilson, 2002). Moreover, recent analyses of prime time television programs do indeed indicate that sexual content is prevalent, appearing in 70% of prime-time television programs (Kunkel et al., 2005), and recent media effects studies find that sexual content is associated with sexual behavior patterns (Brown, L'Engle, Pardun, Guo, Kenneavy, & Jackson, 2006; Collins et al., 2004). Television content may be considered, and analyzed, many ways, depending on the theoretical framework underlying the questions. For example, researchers working from Social Learning Theory (Bandura, 1977) would argue that it is most important to examine the characters with which adolescent audiences identify, as studies have shown that viewers are more likely to imitate a character's behavior if they identify with that character. In this case, a character-based analysis--one that tracks one or more focal teen characters' relationships and sexual behaviors--is appropriate. In addition (or alternatively), one may work from a theoretical perspective that suggests that the prevalence of behaviors and the consequences of those behaviors are important to assess for their potential impact on teen sexuality. A theory such as the Theory of Reasoned Action (Fishbein & Ajzen, 1975) would privilege behavior-level variables as appropriate elements of the program to study.

A key challenge facing researchers interested in the landscape of television programming today is the plethora of programming available. To address this, researchers often focus on a subgroup of programs to examine, for example, shows that are popular with a particular audience (e.g., Byrd-Bredbenner, Finckenor, & Grasso, 2003; Franzini, 2003) or that represent a particular genre, such as soap operas (e.g., Heintz-Knowles, 1996). In addition, researchers may choose some combination of attributes, such as comedies appearing during the prime-time hours (Lampman et al., 2002).

Yet even when researchers have selected a particular show or set of shows to include in their analysis, it often is impractical to content analyze every episode of a program or set of programs, especially for longitudinal studies looking at several seasons of programming, or for studies examining a large sampling frame of program titles. For this reason, researchers generally select a sample of episodes to include in their analysis. In rare instances, researchers will focus on characters and their evolution over the course of a season, necessitating the need to examine each episode of the series (e.g., Franzini, 2003). Most studies focus on behaviors, or incidents of sex that occur in a program (e.g., Lampman et al., 2002). Ultimately, the focus of the analysis may influence the type of sampling necessary.

One could argue that the decisions researchers must make about their sampling may have consequences for findings. Pardun, L'engle, and Brown (2005) examined one episode each of 71 television shows in order to measure adolescent exposure to sexual content. Alternatively, Collins and colleagues (2004) analyzed 2-13 episodes each of 23 different programs and similarly attempted to connect exposure with sexual behavior. While Pardun and her colleagues found no link between exposure to television content and subsequent sexual behavior, Collins' team did find that heavy exposure predicted earlier sexual initiation. The striking difference in choices about how many episodes to analyze highlights the lack of clear direction in basic sampling procedures. In other words, is one episode enough to capture sexual content? Are 13 episodes too many?

Our review of the extant literature on television content analysis reveals that there is no "gold standard" or tested set of practices that has been established to guide researchers, ultimately leaving those who use content analysis research to try to draw conclusions from previously published studies concerning what constitutes an adequate sample size or the most efficacious sampling techniques. There is a variety of approaches one might take, but what information do researchers gain or lose by choosing one particular sampling strategy over another? And how do decisions about sampling ultimately shape findings concerning both messages in the media and media effects related to sexual behavior?

Some progress has been made in answering this question with magazine and broadcast news content. Lacy, Rifle, and Randle (1998) were interested in the sample size needed for monthly consumer magazines. The study involved the analysis of five variables (including number of articles and number of ads) from 5 years of two magazines. After generating 50 different sets of eight random samples (from 6 to 22 issues), they compared the statistics from each sample with the population data collected from the analysis of all magazine issues. This number of random samples (50 per magazine) was chosen to allow for adequate comparison of means between the random and stratified samples, based on the central limits theorem. The central limits theorem posits that "68% of sample means (on a particular variable) should fall within one standard error of the population; 95% should fall within two standard errors" (Lacy et al., 1998, p. 410). The authors concluded that the best way to sample would be to create a constructed year sample by randomly selecting an issue from each month, resulting in a total of 12 magazines to analyze. They argued that this was more efficient than taking a random sample of 20 magazines (20 being the minimum number of issues they determined were required to achieve accuracy based on the above criteria.)

An earlier study used similar methods to examine the best sampling strategy for weekly news magazines, as opposed to monthly magazines (Rifle, Lacy, & Drager, 1996). The authors generated 50 simple random samples using a variety of sample sizes, ranging from six issues to 16 issues, of a year's worth of issues of Time (n = 52). They also generated a sample of 12 issues stratified by month. The authors concluded that when examining content from weekly magazines, the best strategy was to create a constructed year by selecting one issue from every month, for a total of 12 issues.

Finally, another group of authors examined sampling of television broadcast news appearing on weekdays for a year (Rifle, Lacy, Nagovan, & Burkum, 1996). The authors compared eight different simple random samples (ranging from 5 to 40 weekdays), constructed weeks, and stratified monthly sampling. Using a method similar to the studies described above, they selected 40 samples for each sampling strategy, and they compared the results of the samples with the population data. They found that constructing a sample using monthly stratification by randomly selecting 2 days from each month provided a sample most reflective of the universe.

Although a majority of content analysis studies focus on analyzing sexual content from weekly television programs, we know of no study similar to the ones above that systematically has examined whether certain sampling methods are more efficient and effective than others. We argue that such a study is critical, as the results of a content analysis may be different depending on whether one examines the entire season of a program or a randomly selected episode or set of episodes. Understanding the most efficient way to analyze sexual content on television would provide a guide for future research.

Research Questions

This article incorporates data from two separate studies. Study 1 originally was conducted to explore the relationship between television characters' sexual activity and interpersonal success on programming most popular among adolescents. Data collected for Study 1 were used to address the questions of the best sampling strategy for examining character-based sexual content on television. Study 2 was designed as part of a larger study of sexual content to help test codebook questions and to assess the results of different sampling strategies. Study 1 included 10 series examining character-level data, and Study 2 included a single series examining program-level data. These studies focused on different units of analysis (character vs. programs), and offer distinct perspectives for "best practices" in sampling programming to capture sexual content on television. Using data from these two data sets, we aim to answer three main research questions:

* What is the minimum number of episodes one can randomly select from a season to ensure that the selected episode(s) are representative of sexual content from all episodes of an entire season when examining character-level measures? (Study 1)

* What is the minimum number of episodes one can randomly select from a season to ensure that the selected episode(s) are representative of content from all episodes of an entire season when examining program-level measures? (Study 2)

* Are there episodes during the season--specifically the premiere and/or finale episodes--that are significantly different, in terms of sexual content, from other episodes aired throughout the season? (Studies 1 & 2)

Method

Study 1: Sampling for Character-Level Sexual Content

Sample. This study examined all episodes (excluding reruns) of the entire 1998-1999 season of primetime, network shows most viewed by 12-17-year-old males and females (Johnson, 1998). (1) These two lists (male and female) provided a total of 13 fictional (i.e., dramatic or comedic) shows, although 3 of the programs were dropped by the networks prior to the start of the study in the 1998 1999 season, resulting in a total of 10 shows: Beverly Hills, 90210; Boy Meets World," Dawson's Creek," ER; Friends; Home Improvement; King of the Hill," Party of Five; Sabrina, the Teenage Witch; and The Simpsons. (See the Appendix for program information, including networks, viewing times, and number of episodes.)

Unit of analysis. The unit of analysis for Study 1 was the individual character. More specifically, the study analyzed all plot-functional characters who were adolescents or older. A plot-functional character is defined as a character who plays an integral role in the narrative of the show. The character was chosen as the unit of analysis because the original focus of the study was to understand the relationship between the characters' sexual activity and their longer-term (over the course of the season) interpersonal success (happiness and success in their personal and professional lives).

Coding instrument. The coding instrument individually coded each character in each episode where that character appeared. Content-analytic variables were organized around three categories: (1) programming, (2) demography, and (3) sexual behavior. The coding instrument is available upon request.

Programming variables included the program title, the episode number, the network on which the program aired, the program type/genre (live-action sitcom, animated series, or drama), program length (half-hour or hour), and the day of the week and time of night the program originally aired. Demography variables measured characters' gender, race, age, social class, and sexual preference. Sexual behavior pertained to the nature of the portrayal of characters' physical intimacy. This variable was coded using a three-part scale: no sex (no sexual behavior to a simple platonic "peck"), kissing/necking (a nonplatonic kiss through heavy petting), or sexual intercourse (implied or actual sexual intercourse). It is important to note that since the unit of analysis was the character, a character's sexual intimacy would be counted only once during an episode. Therefore, if a character kissed his or her partner three times in one episode, it was only counted once. The highest "level" of sexual intimacy was coded (i.e., if a character kissed a partner and then went on to have sexual intercourse with him or her, then sexual intercourse was coded). (2)

Administration. All 227 episodes of the 10 shows were videotaped. Content was coded by a primary coder. A second person coded 10% of the episodes to establish intercoder reliability, which was measured for each variable by both simple agreement and Cohen's kappa. The numbers ranged from 87% to 100% for simple agreement (average = 99%), and .71 to 1.0 for Cohen's kappa (average = .94), which accounts for agreement by chance. Only variables with a simple agreement of .8 or greater and a kappa of .70 or greater were included in this study. (3)

Study 2: Sampling for Program-Level Sexual Content

Sample. Episodes of the same season of Friends, which aired from 1998 to 1999, were examined for Study 2. There were a total of 23 episodes. While the coding instrument used in Study 1 was designed to examine character-level variables, Study 2 utilized a coding instrument designed to examine program-level behaviors.

Unit of analysis. The unit of analysis for Study 2 was the entire television program.

Coding instrument. Two researchers were trained to code episodes for types of sexual talk and behaviors that occurred in the episodes. Variables were coded dichotomously (present, not present). The coding instrument is available upon request.

Administration. Both coders coded all episodes. The reliability assessment of the variables used in the analyses ranged from 83% to 100% (average = 86%) for simple agreement and .72 to .83 for Cohen's Kappa (average = .80).4 As with Study 1, only variables with a simple agreement of .8 or higher and a kappa of .70 or higher were included in this study.

Sexual behaviors variables included passionate kissing, nudity/partial nudity, and sexual intercourse interrupted. Sexual talk included talk about sexual actions/hooking up, making plans to have sex at a later time, and expert advice about sex/sexual health. These variables represent a range of low to high frequencies across all 23 episodes. This ensured that findings were not specific to variables depending on frequency. Sampling strategies might produce different results if a behavior appeared in all episodes compared with a behavior that occurred in only one program from the season.

Analytic Techniques

In order to assess various sampling strategies, different samples were drawn from the entire selection of all episodes. The sampling strategies were as follows: (1) premiere episode; (2) finale episode; (3) sweeps episodes (episodes aired during November, February, and May, which are considered "sweeps" months; Walker & Ferguson, 1998); (4) fifth episode; (5) 1 randomly selected episode," (6) 3 randomly selected episodes; (7) 5 randomly selected episodes," (8) 7 randomly selected episodes; and (9) all episodes (which ranged from 20 to 26, depending on the number of episodes per program). With respect to randomly selected episodes, 7 was the maximum number we examined, which represents about one third of the episodes that typically appear during a television season.

Random selection of episodes was conducted differently for Study 1 and Study 2. For Study 1, which analyzed 10 programs, 50 random samples were drawn for each of the random sample types (listed above), for all shows combined and for each specific show. For Study 2, which examined a single series (Friends), 50 random samples were drawn for each of the random sample sampling strategies (1, 3, 5, and 7 randomly selected episodes). Then, the average of the resulting variable percentages, standard deviation, and confidence interval were calculated for each random sample type using all 50 samples drawn. This procedure is similar to that used by Stephen Lacy, Daniel Rifle, and colleagues as described in the beginning of this article (Lacy et al., 1998; Rifle, Lacy, & Dranger, 1996).

The SAMPLE command in STATA 8 (StataCorp, 2004) was used to randomly select episodes. Data were analyzed using Microsoft Excel, SPSS Statistical Program for Windows (SPSS, 2002), and STATA 8 (StataCorp, 2004). SPSS and STATA were used to produce frequencies, cross tabulations, and chi-square analyses.

Findings

As seen in Table 1, for the first study looking at character-level variables, there was no statistically significant difference found between the type of sample drawn and the key sexual content variable, sexual intimacy. The percentage of characters having no sex ranges from 85% (premiere episode) to 90% (5 random episodes), while the percentage of characters kissing/ necking (more than platonic kissing through necking) ranges from 5% (premiere episode) to 12% (1 random episode). The percentage of characters having sexual intercourse ranges from 0% (fifth episode) to 10% (premiere episode). None of these differences, however, was statistically significant. In addition, there was no major difference when looking at just the premiere or just the finale episodes.

Table 1 illustrates that when examining character-level content related to sexual intimacy for all 10 shows, the sampling strategies do not yield significantly different findings for sexual content. To determine if the sampling method might bias the results from one particular show, however, we also examined each individual show. These analyses indicate that the sampling method produced significantly different findings for only two shows, Friends and Party of Five. This finding appears to be due to a higher amount of sexual content during the premiere episode (Friends." 33.3% in premiere vs. 9.1% for all episodes; Party of Five: 36.4% in premiere vs. 5.4% for all episodes). Of the entire sample, these two shows accounted for 28% of the sexual intercourse appearances, and 15% of the kissing/necking.

Given the higher frequency of sexual intercourse for the characters on these two programs, the 7 random episodes sample provided a better estimate for the sexual content, as that sampling strategy most closely matched the sampling method that included all episodes of the program.

The second study provided insight into whether there are differences in results from various sampling strategies when comparing findings from character-based sexual content measures, which look at whether the character engaged in a behavior during the show, with program-based measures (whether a behavior occurred at any point in the program). For instance, it could be that a main character in a television program is sexually active, so in every episode this character engages in a sexual act. Thus, for a character-based content analysis, the selection of episodes would not alter the findings. When looking at program-level variables, however, we can see that it is possible that one particular episode could have multiple incidents of sexual activity by the same character while other episodes do not. In this case, a fewer number of episodes may give an inaccurate view of the behavior portrayed across the entire season.

Table 2 presents the results from the various sampling strategies used when examining sexual behaviors and talk appearing in the sitcom Friends. The percentages reflect the average of 50 sets of randomly selected episodes. When using a sample with only one episode, including the 1 random episode sample, for a dichotomous (present/not present) variable, the results are either 100% or 0%. Thus, if nudity occurred in the one episode that was examined, the percentage would 100%. This is misleading, however, since it would seem to indicate that nudity was in every episode.

Table 2 indicates that any of the samples using 1 episode led to an assessment that was not representative of sexual content appearing in the program as a whole, regardless of which episode is selected. These data suggest that a 3 random episode sample provides the best strategy to assess the overall content of an entire season of a show. When looking at the statistics for all 50 samples drawn, analysis showed that the average of all 50 samples for each variable fell within the confidence interval (2 standard errors above or below the mean). This was also true for the 5 random episodes and 7 random episodes samples, but it was not true for the 1 random episode sample, where one variable, sex planned but interrupted, did not fall within the confidence interval. Thus, the 3 random episodes sample was the sample with the lowest number of episodes that led to results representative of the entire season.

Results for the 3 random episodes sample suggest that for one variable, nudity, the first sample drawn gives a much different percentage than the percentage from all episodes. Even with the 5 random episodes samples, however, when looking at individual samples drawn, there is a range of 0% to 60% for nudity and 0% to 80% for passionate kissing. The large range remains even when looking at 7 random episodes samples. For instance, there is a range of 0% to 43% for nudity and 14% to 71% for passionate kissing. As to be expected, when a variable appears very infrequently (close to 0% when examining all episodes) or very frequently (close to 100%), the ranges resulting from each individual sample are smaller and closer to population percentage. For a behavior such as passionate kissing, which varies more across episodes, the range of percentages resulting from each sample type also varies more, and can sometimes fall far from the population percentage.

An additional finding of interest occurred in comparing sexual content from the character-based analysis to the program-based analysis in the program Friends, the program examined in both Study 1 and Study 2. When examining all episodes of Friends, we found that in the character-level analysis, 9% of characters had sexual intercourse, while in the program-level analysis, 22% of episodes had "implied sexual intercourse" in them (no depicted sexual incidents occurred in any episode). Both methods of sampling found 0% for the final episode and the fifth episode. For the sweeps episodes, 2% of characters had implied sexual intercourse, while 10% of episodes had instances of implied sex. The largest difference occurred with the premiere episode. For that single episode, the character analysis determined that 33% of characters had sex, compared with 100% for the program-level analysis (meaning that implied sex occurred at least once on the program). These numbers suggest that very different findings can occur when doing a character-based vs. program-based analysis.

Conclusions

Study 1 results indicate some differences among the various sampling strategies when examining sexual content at the character level. One research question considered the minimum number of episodes needed for character-based measures of sexual content. This study found that of the sampling strategies used, seven random episodes is the strategy most appropriate for character-based measures of sexual content. It is likely that a greater number of episodes may be needed for programs with higher sexual content (Friends, Party of Five), while a fewer number of episodes may be required for programs with less sexual content. Qualifications aside, a sample of seven randomly selected episodes is adequate for character-based measures.

Findings from the study of program-level measures (Study 2) suggest that a different sampling strategy should be used when examining sexual behavior at the program level. When research is conducted using a television program as the unit of analysis, an analysis of five randomly selected episodes presents a valid portrait of the overall season of programming. Prior studies have suggested that high levels of variation may require larger sample sizes (Lacy, Rifle, Stoddard, Martin, & Chang, 2001; Rifle et al., 1996). The data from this study, however, also suggest that randomly selecting just three episodes will provide a reasonable assessment of a series' content if researchers do not expect wide variation. Researchers have argued that the premiere episode is likely to differ from episodes airing throughout the rest of the season because the programs want to capture the interest of viewers so they will continue watching throughout the season (for instance, see sampling description in Lauzen & Dozier, 2002). The data reveal that the premiere episode and the finale episode did differ from the overall content of the programs in Study 1, although the differences were not statistically significant. This study found that the sampling strategy using only the premiere episode resulted in the highest sexual content with respect to the percent of characters having sex (10%), and the second highest of the sampling strategies with respect to sexual opportunity (17.5%). For Study 2, the premiere and finale samples did not seem to differ from the fifth episode sample, but all three of these sampling strategies produced results that are difficult to interpret, since the variable was dichotomous.

Although the random sample recommendations may be a valid method for providing a representative sample of the overall content of a program, one must have either a census from which to draw episodes or an advance calendar of the programming schedule. This is a challenge for television content analysis researchers, as the number of programs in a given season is variable, and air dates often are unpredictable. Taping all episodes of a program (or acquiring the season on DVD) is ideal, although this can add significant time and expense to a research study.

There are some important limitations of this study. All programs examined were programs that appeared once per week. Sampling strategies may differ for programs that are shown daily, such as talk shows, soap operas, or news (Rifle, Lacy, Nagovan, & Burkum, 1996). In addition, this particular study focused on sexual behaviors. Findings may be different for other behaviors or attributes that are examined by content analysis studies. Although we compared findings by counting character-based sex and sexual talk and behavior from one show during the same season, it may be that looking at other shows would lead to different results. Friends, a 30-minute comedy, may be different from other types of shows, such as cartoons or one-hour dramas. Also, the sample for Study 1 only included programs popular with adolescents. The findings may differ when analyzing sampling strategies or sexual content on programming popular among the greater population. Finally, although Friends was a prime-time mainstay for years and still airs in syndication, it is possible that today's programs have distinctly different sexual content patterns.

Despite these limitations, this study offers an important first step in determining the most efficient and accurate methods for drawing a representative sample of episodes in television content analyses focusing on sexual content. The results of this study suggest that a random sample of 3 or 5 episodes from a season of television should provide a reasonable assessment of the overall sexual content of the program for program-based studies, while a random sample of 7 episodes is needed for character-based analyses.

Future research should examine whether sampling strategies may differ for other sexual content, including gender roles and sexual messages (e.g, Ward, 1995), sex the workplace (e.g., Lampman et al., 2002), risk messages about sex (e.g., Kunkel et al., 2005), sexual consequences (e.g., Aubrey, 2004), safe sex (e.g., Lowry & Shidler, 1993), and sexual health and birth control (e.g., Lowry & Towles, 1989). Additional research also should explore how sampling strategies may be different for other health behaviors depicted on television, such as violence (e.g., Signorielli, 2003), injury (e.g., Winston, Woolf, Jordan, & Bhatia, 2000), obesity (e.g., Greenberg, Eastin, Hofschire, Lachlan, & Brownell, 2003), nutrition (Tirodkar & Jain, 2003), and substance use (e.g., Mathios, Avery, & Bisogni, 1998). Finally, examining other types of shows (for example, cartoons, non-teen-oriented television, talk shows) and comparing findings for character coding to incident or behavior coding for more than one show across an entire season also would provide useful information to researchers. Ultimately, of course, all researchers conduct content analyses with the goal of better understanding media impact on audiences. Refining our research methods will bring us closer to that goal.

APPENDIX Television Shows in Study 1 Sample

                                                            #
Name of Show           Network         Day/Time          Episodes

Beverly Hills, 90210     FOX     Wednesday/ 8:00 p.m.       26
Boy Meets World          ABC     Friday/8:30 p.m.           20
Dawson's Creek           WB      Wednesday/8:00 p.m.        22
ER                       NBC     Thursday/10:00 p.m.        22
Friends                  NBC     Thursday/8:00 p.m.         23
Home Improvement         ABC     Tuesday/8:00 p.m.          23
King of the Hill         FOX     Tuesday/8:00 p.m.          24
Party of Five            FOX     Wednesday/9:00 p.m.        24
Sabrina, the Teenage             Friday/8:00 p.m.
  Witch                  ABC       (switched to 9:00)       23
The Simpsons             FOX     Sunday/8:00 p.m.           20

References

Aubrey, J. (2004). Sex and punishment: An examination of sexual consequences and the sexual double standard in teen programming. Sex Roles, 50, 505-514.

Bandura, A. (1977). Social learning theory. Engelwood Cliffs, NJ: Prentice Hall.

Brown, J. D., L'Engle, K. L., Pardun, C. J., Guo, G., Kenneavy, K., & Jackson, C. (2006). Sexy media matter: Exposure to sexual content in music, movies, television, and magazines predicts Black and White adolescents' sexual behavior. Pediatrics, 117, 1018-1027.

Byrd-Bredbenner, C., Finckenor, M., & Grasso, D. (2003). Health related content in prime-time television programming. Journal of Health Communication, 8, 329-341.

Capella, J. N., Price V., & Nir, L. (2002). Argument repertoire as a reliable and valid measure of opinion quality: Electronic dialogue during campaign 2000. Political Communication, 19, 73-93.

Collins, R., Elliott, M., Berry, S., Kanouse, D., Kunkel, D., Hunter, S. B., & Miu, A. (2004). Watching sex on television predicts adolescent initiation of sexual behavior. Pediatrics, 114, 280-289.

Fishbein, M. & Ajzen, I. (1975). Belief attitude, intention & behavior: An introduction to theory and research. Reading, MA: Addison-Wesley.

Franzini, A. (2003, August). Sex on television programming popular among teens: Correlates and consequences. Unpublished doctoral dissertation, Temple University, Philadelphia.

Greenberg, B. S., Eastin, M., Hofschire, L., Lachlan, K., & Brownell, K. D. (2003). Portrayals of overweight and obese individuals on commercial television. American Journal of Public Health, 93, 1342-1348.

Heintz-Knowles, K. E. (1996). Sexual activity on daytime soap operas: A content analysis of five weeks of television programming. [on-line]. Available: http://www.kff.org/entmedia/1190-soap2rpt. cfm

Johnson, T. (1998). The most powerful people in Hollywood. TV Guide, 46, 42-48.

Kunkel, D., Eyal, K., Finnerty, K., Biely, E., & Donnerstein, E. (2005). Sex on TV 4: A biennial report to the kaiser family foundation. Menlo Park, CA: The Kaiser Family Foundation.

Lacy, S., Riffe, D., & Randle, Q. (1998). Sample size in multi-year content analyses of monthly consumer magazines. Journalism & Mass Communication Quarterly, 75, 408-417.

Lacy, S., Riffe, D., Stoddard, S., Martin, H., & Chang, K. (2001). Sample size for newspaper content analysis in multi-year studies. Journalism & Mass Communication Quarterly, 78, 836-845.

Lampman, C., Rolfe-Maloney, B., David, E., Yan, M., McDermott, N., Winters, S., et al. (2002). Messages about sex in the workplace: A content analysis of primetime television. Sexuality & Culture, 6, 3-21.

Lauzen, M. & Dozier, D. (2002). You look mahvelous: An examination of gender and appearance comments in the 1999-2000 prime-time season. Sex Roles, 46, 429-37.

Lombard, M., Snyder-Duch, J., & Bracken, C. C. (2002). Content analysis in mass communication: Assessment and reporting of intercoder reliability. Human Communication Research, 28, 587-604.

Lowry, D. T. & Shidler, J. (1993). Prime time TV portrayals of sex, "safe sex" and AIDS: A longitudinal analysis. Journalism Quarterly, 70, 628-637.

Lowry, D. T. & Towles, D. E. (1989). Prime time television portrayals of sex, contraception, and venereal diseases. Journalism Quarterly, 66, 347-352.

Magzamen, S., Charlesworth, A., & Glantz, S. A. (2001). Print media coverage of California's smokefree bar law. Tobacco Control, 10, 154-160.

Mathios, A., Avery, R., & Bisogni, C. (1998). Alcohol portrayal on prime-time television: Manifest and latent messages. Journal of Studies on Alcohol, 59, 305-310.

Pardun, C. J., L'Engle, K. L., & Brown, J. D. (2005). Linking exposure to outcomes: Early adolescents' consumption of sexual content in six media. Mass Communication & Society, 8, 75-91.

Riffe, D., Lacy, S., & Drager, M. W. (1996). Sample size in content analysis of weekly news magazines. Journalism & Mass Communication Quarterly, 73, 635-644.

Riffe, D., Lacy, S., & Fico, F. G. (1998). Analyzing media messages. Using quantitative content analysis in research. Mahwah, NJ: Lawrence Erlbaum Associates.

Riffe, D., Lacy, S., Nagovan, J., & Burkum, L. (1996). The effectiveness of simple and stratified random sampling in broadcast news content analysis. Journalism & Mass Communication Quarterly, 73, 158-168.

Signorielli, N. (2003). Prime-time violence 1993-2001: Has the picture really changed? Journal of Broadcasting & Electronic Media, 47, 36-57.

SPSS Inc. (2002). SPSS for Windows (Release 11.5.1.) [Computer Software] Chicago, IL: Author.

StataCorp LP. (2004). Intercooled Stata 8.2 for Windows [Computer Software]. College Station, TX.

Strasburger, V. & Wilson, B. (2002). Children, adolescents, & the media. Thousand Oaks, CA: Sage Publications.

Tirodkar, M. & Jain, A. (2003). Food messages on African-American television shows. American Journal of Public Health, 93, 439-441.

Walker, J. & Ferguson, D. (1998). The broadcast television industry. Boston, MA: Allyn & Bacon.

Ward, L. (1995). Talking about sex: Common themes about sexuality in the prime-time television programs children and adolescents view most. Journal of Youth and Adolescence, 24, 595-615.

Winston, F. K., Woolf, K. D., Jordan, A., & Bhatia, E. (2000). Actions without consequences: Injury-related messages in children's programs. Archives of Pediatrics & Adolescent Medicine, 154, 366-369.

(1) Although the overall Nielsen rankings of television shows routinely are reported, specific audience demographics typically are reserved for paying clients and, thus, not easily available for general academic research. In the September 26, 1998, issue of TV Guide, however, particular attention was focused on teenagers, and with that, listings of both the top 10 shows for 12-17-year-old females and males were published.

(2) There were no instances of characters engaging in any level of sexual intimacy with more than one partner in an episode.

(3) Although a coefficient of .8 or higher is the recommended standard in content analysis research (Lombard, Snyder-Duch, Bracken, 2002; Rifle, Lacy, & Fico, 1998), many researchers treat .7 as the practical standard, especially when simple agreement is high (i.e., Capella, Price, & Nir, 2002; Magzamen, Charlesworth, & Glantz, 2001).

(4) The two coders compared responses after coding, and when different answers were marked, the coders came to agreement on what the "correct" answer should be. In cases where the coders could not come to agreement, a third person broke the tie. The data used for this analysis reflect the agreed-upon set of answers.

Jennifer Manganello

School of Public Health, University at Albany

Amy Franzini

Department of Communication Studies, Widener University 

Amy Jordan

Annenberg Public Policy Center, University of Pennsylvania 

Correspondence should be addressed to Dr. Jennifer Manganello, Department of Health Policy Management and Behavior, School of Public Health, University at Albany, One University Place, Rensselaer, NY 12144-3456. E-mail: [e-mail]

Table 1. Study 1, Cross Tabulations of Sexual Intimacy by Sample Type
for Character-Level Analysis for Top 10 Teen-Oriented Programs

                                         Kissing
                           No Sex       /Necking

                           N     %      N      %

Premiere                    68   85       4     5
All episodes             1,431   86     135     8
[x.sup.2] = 3.782,
  d .f. = 2, p = .151
Finale                      72   87       7     8
All episodes             1,431   86     135     8
[x.sup.2] = .064, d.f.
   = 2, p = .968
Sweeps                     629   86      58     8
All episodes             1,431   86     135     8
[x.sup.2] = .219, d.f.
   = 2, p = .896
5th episode                 66   89       8    11
All episodes             1,431   86     135     8
[x.sup.2] = 4.691,
  d.f. = 2, p = .096
1 Random                    57   85       8    12
All episodes             1,431   86     135     8
[x.sup.2] = 31.850,
  d.f. = 2, p = .396
3 Random                   191   88      14     7
All episodes             1,431   86     135     8
[x.sup.2] = .801, d.f.
  = 2, p = .670
5 Random                   333   90      24     7
All episodes             1,431   86     135     8
[x.sup.2] = 4.407,
  d.f. = 2, p = .110
7 Random                   468   86      43     8
All episodes             1,431   86     135     8
[x.sup.2] = .185, d.f.
  = 2, p = .912

                           Sexual
                         Intercourse     Total

                           N     %      N      %

Premiere                     8   10      80   100
All episodes                90    5   1,656    99
[x.sup.2] = 3.782,
  d .f. = 2, p = .151
Finale                       4    5      83   100
All episodes                90    5   1,656    99
[x.sup.2] = .064, d.f.
   = 2, p = .968
Sweeps                      43    6     629   100
All episodes                90    5   1,656    99
[x.sup.2] = .219, d.f.
   = 2, p = .896
5th episode                  0    0      66   100
All episodes                90    5   1,656    99
[x.sup.2] = 4.691,
  d.f. = 2, p = .096
1 Random                     2    3      67   100
All episodes                90    5   1,656    99
[x.sup.2] = 31.850,
  d.f. = 2, p = .396
3 Random                    11    5     216   100
All episodes                90    5   1,656    99
[x.sup.2] = .801, d.f.
  = 2, p = .670
5 Random                    12    3     369   100
All episodes                90    5   1,656    99
[x.sup.2] = 4.407,
  d.f. = 2, p = .110
7 Random                    32    6     543   100
All episodes                90    5   1,656    99
[x.sup.2] = .185, d.f.
  = 2, p = .912

Percentage of all episodes does not equal 100 due to rounding.

Table 2. Study 2, Percent of Each Variable Occurring in the Episode(s)
by Type of Sample Drawn from the 1997-1998 Season of Friends

                  Passionate       Sex Planned     Nudity/Partial
                   Kissing       but Interrupted       Nudity
                      %                 %                %

Premiere             100              100                 0
Finale               100                0               100
Sweeps (N,F,M)        30                0                20
5th Episode          100              100                 0
1 Random *            36                8                22
3 Random *            32               19                14
5 Random *            35               15                18
7 Random *            38               19                20
All episodes          35               17                17

                  Talk about      Making Plans    Expert Advice
                Sexual Activity    to have Sex      about Sex
                      %                 %               %

Premiere             100              100               0
Finale                 0                0               0
Sweeps (N,F,M)        60               20               0
5th Episode            0                0             100
1 Random *            62               12               6
3 Random *            63               13              11
5 Random *            64               20               9
7 Random *            65               21              11
All episodes          65               17               9

* This percentage represents the average of all 50 random samples drawn
for each sampling strategy.