The NSFG is a nationally representative survey of women and men ages 15–49 in the U.S. household population. Female interviewers use computer-assisted interviewing techniques to collect data on a variety of topics, including sexual behavior and contraceptive use [1]. We used the 2011–2013 and 2013–2015 female respondent files of women ages 15–44, including 11,300 women (5601 in 2011–2013 and 5699 in 2013–2015, with respective response rates of 73% and 71%).

We used two sets of recode variables, contraceptive status during the interview month, (referred to in analysis as “recent method use,” recode variables CONSTAT1–4), and contraceptive method used at last intercourse in the past 12 months (referred to in analysis as “method used at last sex,” recode variables METH12M1–4), each consisting of four variables to account for reporting of up to four different methods. For each set of recodes, we first generated one estimate of contraceptive use among a population of women we defined as at risk for unintended pregnancy (referred to as our standard estimate); we then generated a second estimate using additional information to further align the timing of contraceptive use measurement with risk for unintended pregnancy (referred to as our augmented estimate). The next section describes how the measures were constructed.

The current contraceptive status recode variables pull information from throughout the NSFG. Female respondents use a life history calendar to recall monthly sexual activity and up to four contraceptive methods used at any point during each month, going back to January, 3 years before the interview. As described in the specifications for this recode variable, methods reported in the month of interview are priority coded by ranking of effectiveness at preventing pregnancy [12] to define current contraceptive status. The first variable in this set of recodes (CONSTAT1) represents the method used that is most effective method at preventing a pregnancy, unless information obtained elsewhere in the survey indicates that a woman was pregnant or she or her husband or cohabiting partner were infecund for reasons other than surgical sterility, in which case this latter information is given priority [12]. Sterility information for a husband or cohabiting partner is taken from a section of the questionnaire for married and cohabiting women, but women with other types of sexual partners could report on their partner’s sterility as part of the life history calendar, which is included in the construction of the CONSTAT recodes [12]. For women who report multiple methods in the interview month, the remaining methods are then assigned in order of effectiveness in the variables CONSTAT2–4. Finally, for women who report no method use during the interview month (and are not pregnant or infecund based on other survey questions asking about their ability to become pregnant and carry a pregnancy to term), the value for CONSTAT1 is assigned to a “non-user” category, in the following ordering: seeking pregnancy; postpartum (completed pregnancy ending less than or equal to two and a half months before the interview); never had intercourse; had intercourse but not at any point during the 3 months prior to interview; and had intercourse during the three months prior to interview [12].

Using the CONSTAT recode series, we created two measures of recent contraceptive use. For our first estimate - standard measure of recent method use - we defined women in our population as at risk for unintended pregnancy if they had intercourse in the last 12 months (based on the recode NOSEX12) and were not currently pregnant, seeking pregnancy, or postpartum, and if they and their partner were not non-surgically sterile (based on the CONSTAT1 recode). For this estimate, we did not differentiate women based on whether they had intercourse during their interview month. Fig. 1a describes the analytic population (n = 7498).

For our second estimate of recent contraceptive use - the augmented estimate - we used the same sub-population of 7498 women from our first estimate (Fig. 1a). However, we incorporated additional information for women who did not use contraception during their interview month (n = 1349). Among these women who did not use a method in their interview month, we used the monthly intercourse variables to identify those who also did not have intercourse that month (n = 947). For these women we used the life history calendar to “look back” and identify the last month the woman reported having intercourse in the past 12 months and then determine the method(s) used during that month (if any). Among this subpopulation of the 947 women, 758 (80%) had their method re-assigned based on what method they used during the last month they had sex; among women re-assigned, for roughly one-third we used information from the month before the interview month, for two-thirds within 3 months of looking back, and for four-fifths within 6 months. Although we looked back to identify the last month the woman had intercourse for women who did not use a method and did not have sex in the interview month, we did not look back to align sexual activity and method use for women who used a method in the interview month but did not have sex (n = 880). Most of these women (99%) were relying on contraceptive sterilization, long-acting reversible contraception, or hormonal contraceptives rather than coital dependent methods.

In the NSFG, contraceptive use is also assessed by asking women who reported at least one male partner in the last 12 months which contraceptive method(s) (if any) they used at last intercourse. This differs from the current contraceptive use or the “recent method” (CONSTAT recode) because it is method use at the time of last sexual intercourse not at any time during a specific month regardless of sexual activity. For the third contraceptive use estimate we present - standard method use at last sex - we used the recode series of method use at last sex (METH12M1–4), selecting the method most effective for preventing pregnancy the women mentioned. The sample size was 8682 women (Fig. 1b).

While the METH12M1–4 recode variables precisely align the timing of contraceptive use with sexual activity, they do not incorporate information from other parts of the interview related to unintended pregnancy risk (e.g., pregnancy status). For the augmented method use at last sex, we used information from women’s pregnancy history to compare the month of last sex to the most recent pregnancy start and end dates; women who were pregnant in the month they last had sex (n = 637) were then excluded from our population at risk for unintended pregnancy. Our final sample for this augmented estimate included 8045 (Fig. 1b).

For all four measures of contraceptive use, we classified methods used into four effectiveness categories: most effective methods (sterilization [male or female], intrauterine device [IUD], or implant); moderately effective methods (injectables, pills, patches, or ring); less effective methods (condoms, withdrawal, or other methods including diaphragm); and no method [19]. We then calculated the percentage and standard error (SE) for each effectiveness category and selected specific methods, overall and by 5-year age groups. We used the sampling weight for the 2011–2015 data and the stratum and cluster variables in SAS SURVEY-FREQ to produce percentages and standard errors. We conducted two-tailed independent samples t-tests (without adjustments for multiple comparisons) to compare the percentages for each standard estimate versus the corresponding augmented estimate, as well as the augmented estimates against each other. We calculated p-values based on these t-tests to check for statistically significant differences. We specified the degrees of freedom as the number of sample clusters minus strata in our calculation of p-values [20].

Using our standard measure of recent method use, we estimated 15.4% of the 41.8 million women ages 15–44 at risk for an unintended pregnancy in the US during 2011–2015 (approximately 6.4 million women) did not use a method of contraception during the month of interview (Table 1). The augmented measure that uses additional information to align contraceptive use and sexual activity for women who did not have sexual intercourse in the interview month, nearly halved this estimate to 7.3% (approximately 3.1 million women) (p < .001). By contrast, our estimate of less effective method use increased from 21.4% using the standard measure to 28.3% using the augmented measure (p < .001); a similar pattern of increase was observed for specific methods within this effectiveness category, including condom use (13.5% vs. 18.7%, p < .001) and withdrawal (6.6% vs. 8.1%, p = .02).

Differences between the standard and augmented recent method measures varied across age groups and were most pronounced for women ages 15–19 (Table 2). Using our standard measure of recent method use, we estimated 29.3% of the 3.4 million women ages 15–19 at risk for an unintended pregnancy did not use a method of contraception (just under 1 million teens); with the augmented measure, this percentage decreased to 5.8% (approximately 200,000 teens; p < .001; Table 2). By contrast, estimates for condom use increased from 18.1% using the standard measure to 35.6% using the augmented measure (p < .001).

For all other age groups, differences between the standard and augmented measures were smaller (Table 2). For no method use, the percentage point decrease from the standard to the augmented measure ranged from approximately 11 (16.7% vs. 6.0%, p < .001) for women ages 20–24 to approximately four for women ages 40–44 (13.8% vs. 9.6%, p = .03). By contrast, the percentage point increase in the use of condoms ranged from approximately seven in women ages 20–24 (16.6% vs. 23.6%, p = .005) to approximately two (8.2% vs. 10.0%, not significantly different) in women ages 40–44.

Using our standard measure of method use at last sex, we estimated 21.0% of women ages 15–44 at risk for unintended pregnancy did not use a method; the augmented measure reduced this estimate to 15.8% (p < .001; Table 1). By contrast, estimates of most effective method use increased from 31.1% using the standard measure to 33.3% using the augmented measure that excluded women who were pregnant (p = .04).

Differences between standard and augmented estimates of method use at last sex again varied across age groups (Table 2). For no method use, the percentage point decrease from the standard to the augmented measure ranged from a high of approximately eight percentage points for women ages 25–29 (24.4% vs. 16.0%, p < .001) and 30–34 (25.5% vs. 17.4%, p < .001), to a low of less than one percentage point for women ages 40–44 (17.7% vs. 17.1%, not significantly different). The difference in no method use for women ages 15–19 was approximately four percentage points (12.6% vs. 8.4%, p = .04). The difference in no method use for women ages 20–24 was approximately five percentage points (19.1% vs. 13.7%, p = .002). There was no statistically significant difference between the standard and augmented estimates of no method use for women ages 35–39. There were no statistically significant differences between the standard and augmented estimates of method use at last sex for other methods or method groupings.

Estimates of all women ages 15–44 at risk for unintended pregnancy not using any method were significantly lower using the augmented measure of recent method use (7.3%) as compared to the augmented measure of method use at last sex (15.8%; p < .001) (Table 1). In addition, estimates of women using the most effective methods were higher using our augmented measure of recent method use (37.5%) as compared to our augmented measure of method use at last sex (33.3%; p < .001). Similarly, estimates of women using sterilization were higher using our augmented measure of recent method use (26.2%) as compared to our augmented measure of method use at last sex (23.1%; p = .01).

Among teens ages 15–19, estimates of no method use did not differ significantly for the two augmented measures, but in all older age groups estimates were lower using the augmented measure of recent method use versus method use at last sex, with an approximately seven to 11 (p < .001) percentage point difference (Table 2). For most effective method use, our estimates were similar for the two augmented measures among women in younger age groups (15–29 years), but there was an increase of approximately six to eight percentage points from our augmented measure of method use at last sex to our augmented measure of recent use among women in older age groups (30–44 years; p < .05).