We analyzed data from the NSFG, a nationally representative household survey of the non-institutionalized civilian population of US women and menFootnote 1 aged 15–44 in the USA (Lepkowski et al., 2010). We pooled data from the 2006–2010 and 2011–2015 waves for a total sample size of 23,579 women and 19,724 men; we then used restricted data on respondents’ state of residence, accessed through the National Center for Health Statistics Research Data Center, to merge the pooled data with state-level measures of abortion stigma developed for this analysis (described below).
The NSFG asked all respondents to report pregnancies first in the FTF interview and then again in the ACASI portion. In the FTF interview, female respondents were asked how many times they have ever been pregnant, followed by detailed questions about each pregnancy, including the pregnancy outcome. In the ACASI they were asked separately their number of live births, abortions and miscarriages within the last five calendar years. Both the FTF and ACASI modes asked male respondents the total number of pregnancies they had ever fathered that did not end in a live birth and the outcome of those pregnancies.
Informed by theoretical literature, we developed five proxy measures of different aspects of structural stigma (identified in italics below). First, we used a variable of the state policy environment, constructed in previous work, that characterized states as hostile to abortion, supportive, or ‘middle-ground’ based on the number of restrictions in the state in 2011 (Gold & Nash, 2012).
We also obtained proxy measures of the visibility of abortion within the state, which could influence how common or accepted respondents feel the procedure is: the abortion rate per 1000 women of reproductive age by state of residence (averaged across 2008 and 2013) and the number of abortion clinics per 1000 women of reproductive age (averaged across 2008 and 2014, and defined as facilities providing 400 or more abortions per year, which provide the large majority of abortions in the USA, and which are likely the most visible to respondents in our sample; Jones & Jerman, 2017).
We used data from the 2014 Religious Landscape Study to construct two distinct measures of the state cultural environment that may influence abortion stigma (Pew Research Center, 2016). First, to measure public opinion about abortion we categorized states by the proportion of respondents that believed abortion should be legal in “all or most cases.” The second was a measure of state religiosity (the proportion of respondents that said religion was “very important” in their life); while not all religions proscribe abortion, the links between religious belief and abortion stigma are well-documented (Frohwirth et al., 2018), and it provides a reasonable proxy measure for religion-based stigma.
The individual-level analysis controlled for sociodemographic characteristics measured in the NSFG that were associated with abortion underreporting in prior studies: age at time of interview (15–24, 25–29, 30–34, 35 +), race and Hispanic ethnicity (non-Hispanic white, non-Hispanic black, non-Hispanic other, Hispanic), parity (0 or 1 + prior births), household income at time of interview (0–99% of the federal poverty line, 100–299%, and 300% +), a dichotomous measure of urban residence, informal marital status (married, cohabiting, not in union), nativity (US-born or foreign-born), and a dichotomous indicator of whether the respondent said religion was very important in their life.
We conducted two interrelated analyses of abortion reporting in the NSFG. The first, among female respondents only, compares weighted aggregate reports of abortions within specific groupings of states with external abortion counts by state of residence derived from censuses of abortion providers in the USAFootnote 2 (Guttmacher Institute, 2018; Jones & Jerman, 2017); the latter are considered the most complete abortion counts available, as not all states report data to the Centers for Disease Control and Prevention (Jatlaoui, 2018).
We grouped states using each of the five proxy measures of abortion stigma (policy environment, abortion rate, number of abortion clinics, public opinion and state religiosity). For each except the first measure, we categorized states into low, middle and high terciles, so that a third of states fell into each category (see Table 1).
Within each state tercile group, we compared the number of abortions in the five years prior to the interview that women reported in the FTF interview with external estimates of the number of abortions obtained by residents of those same pooled states. We adjusted these external estimates to match the NSFG’s sampling frame and five year recall period following an approach used previously for national estimates (Lindberg et al., 2020). Our outcome measure—the level of underreporting—was the proportion of abortions reported in the NSFG as compared to external counts in each state tercile group. We assessed statistical significance on the basis of non-overlapping confidence intervals between state groups, a relatively conservative approach.
We complemented this aggregate analysis with an individual-level analysis of discrepancies in abortion reporting between the FTF and ACASI portions of the questionnaire. Among women and men reporting any abortions in the ACASI portion of the NSFG (men = 2442, women = 1408), we identified if respondents had reported fewer abortions in the FTF interview.For female respondents, we restricted the FTF counts to abortions ending in the five calendar years prior to the interview to parallel ACASI reports, including a buffer period of 6 months on either side to account for slight misdating; for male respondents the counts referred to abortions over their lifetime.Footnote 3 For each gender we estimated separate logistic regressions of the association of reduced abortion reporting in the FTF than ACASI mode with each of the five structural stigma measures for the state in which they reside.Footnote 4 An advantage of this individual-level approach (as compared to the aggregate analysis) is that it allows controls for demographic characteristics known to be associated with abortion reporting; it also allows us to examine underreporting among men, for whom there is no reliable external count of abortions. All analyses accounted for the complex survey design of the NSFG using the svy commands in Stata 15.1 (StataCorp 2017). Note that due to data restrictions, we are unable to include unweighted sample sizes for specific geographic groupings in the tables presented.