We present estimates of age-specific heterosexual partnership formation and dissolution rates using data from a RDD survey of Seattle adults ages 18-39. Using a combination of observed and imputed data, partnership formation rates peaked early – 0.9 per year at age 19 for women and 1.4 per year at age 20 for men – and declined steadily through the 20s and early 30s. Formation of new partnerships among persons with extant partnerships was approximately 20% of the rate among persons without extant partnerships. Over half of all partnerships lasted <=12 weeks and nearly three-quarters dissolved within one year. By the time participants reached their early 30s, >50% had not formed a new sexual partnership in the preceding 3 years, suggesting that a large proportion of the heterosexual population had likely exited the pool of persons at substantial risk for STI by their late 20s (assuming their partners were monogamous).
Our partnership formation estimates fall within the range of existing estimates. Using population-based surveys of Norwegian adults and methods relatively similar to ours, Stigum et al.
estimated that 18-25 and 26-35 year olds had 0.77 and 0.31 new partnerships a year, respectively (14
). Using the same age groups as Stigum et al.
, our observed-plus-imputed findings are somewhat higher – 0.86 and 0.41 – and our observed-only findings are similar. Kretzschmar et al.
used data from a national survey of Dutch adults and estimated a partnership formation rate of approximately 0.63 new partners per year, which is very similar to our overall rate of 0.54 (5
). In contrast, Williams et al.
used data from a survey of U.S. STD clinic attendees (15
) and the National Health and Social Life Survey (4
) and estimated that among persons age 20-24, male and female partnership formation rates were 2.32 and 2.01, respectively (6
). The substantially higher rates found in the latter study may be a result of differences in the population studied. However, we think it is more likely that these differences stem from the assumption that all partnerships within a defined period were newly formed, despite the fact that some certainly formed prior to the reference period.
There are fewer published estimates of partnership dissolution rates. Kretzschmar et al.
estimated that steady partnerships (lasting >=1 year), had a 0.0004 per day probability of ending; casual partnerships (lasting <1 year) had a 0.1 per day dissolution probability (16
). Using similar methods, we found substantially lower dissolution probabilities of 0.0001 per day for steady partnerships and 0.035 per day for casual partnerships.
We presented most findings using both the observed and imputed data to highlight the different estimates provided by the two methodologies. As expected, the observed-only data result in longer partnership durations compared to the observed-plus-imputed analyses. Restricting our analyses to only the observed partnerships results in left-truncated data, where short partnerships which may have begun at the same time as longer partnerships are not captured in the survey. In other words, early partnerships are preferentially excluded from the observed-only data. The observed-only analyses are similar to published estimates, but the observed-plus-imputed methodology may be superior since it gives weight to highly sexually active individuals commensurate to their contribution to the total pool of partnerships in the population.
The observed-plus-imputed analyses have limitations. Longer duration partnerships are most likely to be included in surveys that collect partner-specific data. Thus, the age-specific pools of partnership durations from which we sampled may have been biased resulting in inflated partnership durations, particularly for partnerships initiated at young ages. However, for our imputations we sampled conditionally on the maximum possible duration, which should help minimize this bias. Likewise, our estimates of partnership formation rates in extant partnerships may be underestimates given that these analyses only included the observed data, which tend to oversample longer, steady partnerships. Also, we assumed that “missing” partnerships were equally distributed between sexual debut and the age at the earliest described partnership. Since our findings suggest that partnership formation rates peak early in a person's lifetime, our estimates for younger ages may underestimate the true rates. Indeed, compared to the primary findings, sensitivity analyses that redistributed (backloaded) the “missing” partnerships accentuate the peak formation rate at ages <20 and subsequent decline. Finally, to construct confidence intervals for our observed-plus-imputed estimates, we bootstrapped a dataset following a single imputation, which may have underestimated the variability of the estimates.
This study had other limitations. First, at the time of the survey, 5% of households were cell phone-only households, while 2% lacked any phone, and thus could not have been included in this study (17
). Second, like most previous estimates, we used data from a cross-sectional survey. However, we used a participant's lifetime number of partners to derive our estimates, rather than just current or recent partners. Furthermore, our methods assume that the characteristics (e.g., durations) of current and recent partnerships reflect past sexual norms, although this limitation may be mitigated by the relatively narrow age range of participants. The ideal method for calculating partnership change rates would be to enroll a cohort of participants and prospectively enumerate the formation and dissolution of all partnerships during a relatively long follow-up, preferably a lifetime. Obviously, such a study would be difficult to conduct and might provide data of limited value if sexual behavior changed over calendar time. Third, we did not have data for partnerships formed after age 39. Given the current confluence of longer life expectancy, high divorce rates, and pharmacological treatment for erectile dysfunction, partnership formation rates may increase at older ages. Finally, these data did not provide a large enough sample size to calculate similar rates among homosexual or bisexual men and women.
Defining how partnership formation rates vary over time and differ within and between populations is critical to understanding and modeling of STI transmission dynamics (3
). Our findings provide quantitative estimates of partnership formation and dissolution rates, including rates of concurrent partnership formation and dissolution. They highlight the dramatic extent to which high partner change is concentrated among the young, how a large proportion of partnerships include transient periods of concurrency, and how a majority of the heterosexual population, at least in Seattle, are probably at risk for STI during a relatively short period of their lives. The implications of these findings will be best understood as our results are incorporated into mathematical models of STI transmission. However, more immediately, they suggest that vaccines for STI may not need to induce very long lasting immunity, since most people remain at risk for a brief period of their lives (18
). Our work also highlights the need to collect better, more comprehensive data on the long-term partnership formation patterns of diverse populations – including MSM and racial and ethnic minorities in the U.S. – in order to better explain the profound disparities observed in the occurrence of STI.