The quality of women’s reports of induced abortion is important for improving estimates of the total number of pregnancies, which is essential for demographic and epidemiological studies of the determinants of fertility in populations. It is also important for minimizing bias in the analysis of unplanned pregnancies and contraceptive failures, as well as for the study of induced abortion itself (e.g. the circumstances of recourse to abortion and the conditions of access to it) [12, 18, 20].

Numerous studies have tried to estimate the reliability of data on induced abortion and to characterize the factors associated with underreporting [1, 6, 10, 12, 13, 22]. The main aim of these studies has been to develop survey instruments more favourable to the reporting of induced abortion (questionnaires and interviewing techniques), and to improve ways of controlling for the bias that underreporting introduces into analysis.

Most of these studies involve comparing the rate of induced abortion reported by participants with a rate calculated from the in theory systematic records of induced abortions supplied by health care professionals [6, 10, 13]. A few surveys conducted in the United States [12, 22] and northern Europe [1] have made it possible to study the concordance between women’s reports and the information on the subject collected from their medical records. These studies show that underreporting has many causes : it depends on the socio-demographic characteristics of the women, their experience and their attitudes (reproductive history, views on motherhood, general attitude towards induced abortion etc.). For example, in the United States, being of Afro-American origin, aged over 40, of low educational level and unmarried are all associated with higher underreporting of induced abortion. Reporting is also affected by the survey context (telephone or face-to-face interview, sex and training of interviewer, wording of questions, length of questionnaire etc.) [1, 4, 6, 12, 18, 22].

Some thought also needs to be given to the vocabulary used for talking about induced abortion in questionnaires aimed at the general population. The questions asked almost exclusively use the terms “induced abortion” or “termination” and this can lead to comprehension problems, as emerged from a recent qualitative study on unplanned pregnancies in France, which showed that women used very different terms to describe induced abortion [2]. For although induced abortion is a simple and well-defined act in the medical and scientific literature, its formulation in the general population is much less straightforward, with potential for confusion over many terms. Women who speak of induced abortion or of termination induced by a doctor may in fact be referring to another obstetrical event such as miscarriage (or spontaneous abortion) or therapeutic abortion. Conversely, some women may report a specific obstetrical event when they are really talking about induced abortion. The obstetric terms the most frequently confused with induced abortion are : therapeutic abortion, ectopic pregnancy, spontaneous miscarriage, or miscarriage induced by a doctor (often corresponding in fact to removal of a blighted ovum) [1, 19].

The special attention given in the COCON survey to elaborating the questions relative to induced abortion made it possible for the first time to test the effect of differences in the wording of questions in the same survey. The research had two objectives. The first was to test the hypothesis that the use of questions with less normative formulations than those usually employed would make it easier for women to report this event in quantitative population-based surveys [18, 23]. The second objective was to estimate the classification errors attributable to confusion over vocabulary deriving from the terms “induced abortion” or “termination”, which are those most frequently used in such surveys.

The present study is based on the first wave of the survey using a representative sample of 2,863 women aged 18-44 living in France, interviewed by telephone between September 2000 and January 2001. The women were selected using a complex sampling scheme designed to over-represent those who had had an induced abortion or an unplanned pregnancy in the last five years. The analyses take into account these differential selection probabilities. The results presented here systematically give the crude numbers and the weighted percentages. The survey methods are described in detail in the introductory article of this special issue.

The questionnaire contained four types of question to identify women who had had an induced abortion (see Appendix). The wording of the questions was based on observations from a qualitative study on unplanned pregnancy and induced abortion [2]. These questions were tested in the pilot survey carried out in 1999, particularly as regards the desired legitimizing effect, i.e. the use of language that is less normative than is generally the case [10]. The first questions (questions A in Appendix) were direct : “Have you had an induced abortion or a termination in the last five years ?” or “In the last five years, has a doctor given you a drug called mifepristone whilst you were pregnant ?”. The second set of questions (B in Appendix) sought to legitimize the decision to have an abortion, by directly asking the women how old they were at the time of their first abortion : “Lots of women in France end their pregnancies by recourse to an induced abortion (termination), either by taking drugs such as mifepristone or by undergoing a surgical intervention. If you have already been in this situation, how old were you the first time ?”. The third series of questions concerned the women’s reproductive history (C in Appendix) and elicited the outcome of each reported pregnancy : "The nth time that you were pregnant, did the pregnancy end in : (description of the different possible outcomes including induced abortion) ?”. Finally, a fourth approach, using the so-called ”follow-up” questions, was designed to identify any abortions reported at the beginning of the questionnaire (concerning the previous five years) that were not subsequently described in the reproductive history, by means of the question : ”Is the induced abortion or termination that you mentioned earlier for a different pregnancy to the one you have already told me about ?".

a) Effect of question wording on reporting of induced abortion in the COCON survey

To test the effect of the different wordings used in COCON to identify induced abortions, we compared the lifetime abortion rate obtained from the question “How old were you when you had your first abortion” (question B in Appendix) with that obtained from the responses “induced abortion” or “termination” in the reproductive history section alone (questions C in Appendix) or after correction for the induced abortions traced through the follow-up questions (questions D in Appendix). The first questions (questions A in Appendix) only concerned abortions in the last five years and could not be used to estimate a lifetime abortion rate. Hence it was not possible to compare the results from these questions with those obtained using the other wordings.

b) Effect of question wording on reporting of induced abortions in three different surveys

To test the hypothesis that the use of less normative language facilitates the reporting of induced abortions by women, we compared the rate as estimated from question B of Appendix — “Lots of women in France end their pregnancies by recourse to an induced abortion (termination), either by taking drugs such as mifepristone or by undergoing a surgical intervention. If you have already been in this situation, how old were you the first time ?” — with the rates observed in other surveys on the general population over the same period (KABP survey (Knowledge Attitude Behaviour Practice) [7] and Baromètre Santé (Health Barometer) [8]), which asked the same question more directly : “Have you ever had an abortion, either by taking mifepristone or by surgical means ?” (Health Barometer) or “Have you ever had an abortion ?” (KABP). These surveys used a similar methodology to the COCON survey : interviewing was by telephone on representative samples of the French population (sampling frame based on telephone subscribers, similar to that used in COCON). We compared a subsample of sexually active women aged 18-34, as this was the largest group common to all three surveys.

c) Question comprehension and classification errors

To estimate the classification errors due to confusion of vocabulary over the terms “induced abortion” or “termination”, we drew on the information collected in response to several questions that used different wordings to denote induced abortion and that were asked at different points in the questionnaire. In particular, we used the questions described in part C of Appendix, which were asked for each of the woman’s pregnancies. This allowed us to reclassify each pregnancy outcome as “induced abortion” or “other outcome” and thus to produce an adjusted estimate of the proportion of women who have had at least one induced abortion.

This reclassification also allowed us to measure the “final” size of underreporting in the COCON survey by comparing the estimated abortion rate thus obtained with the national statistics derived from compulsory abortion reports. The comparison was made for 1997, which is the most recent year for which national statistics are available. To improve comparability, the reference rates used for 1997 (based on national statistics) concerned women aged 15-41, which corresponds to the age of the women interviewed in COCON (women aged 18-44 in 2000, i.e. aged 15-41 in 1997).

Table 1
Lifetime induced abortion rates according to the different wordings of the questions in the COCON survey and after reclassification. Comparison with the estimates from the KABP and Health Barometer surveys

The lifetime abortion rates obtained in the COCON, KABP and Health Barometer surveys were broadly similar. The lifetime rate for sexually active women aged 18-34 was 10.2% [8.7-11.7] in COCON according to the question on age at first abortion, and 12.3% in the other two surveys (Table 1). Thus, the less normative language used for questions in the COCON survey seems not to have improved the reporting rate.

The reclassification based on this first question, using the information from the reproductive history and the follow-up questions, is summarized in Table 2, and the criteria used are described in Table 3.

Table 2
Comparison of the responses to the question on age at first induced abortion and the determination of pregnancy outcomes after reclassification

Table 3
Detailed presentation of reclassification of the different pregnancy outcomes according to responses to the question on age at first induced abortion and to the reproductive history (including follow-up questions)

Most of the reclassifications concerned abortions reported in response to the first question (indicated by the single term “induced abortion or termination”) but which were described as another obstetric event in the reproductive history section, rather than a belated identification of abortions that were not reported at the beginning of the questionnaire. Of the 724 women who reported an induced abortion at the beginning of the questionnaire, 48 (6.6%) described another obstetric event in their reproductive history, whereas 13 of the 2,139 women (0.6%) who initially did not report an age at first abortion described this event in their reproductive history.

Compared with the totality of questions that can be used to identify women who have had recourse to abortion, the positive predictive value (percentage of true positives among positive answers to the question on age at first induced abortion) of questions worded using only the terms “induced abortion or termination” was about 93% (676/724) and its negative predictive value (percentage of true negatives among women who responded negatively to this question) was 99% (2,126/2,139).

In all, after reclassification, 689 women aged 18-44 were considered to have had at least one induced abortion in their lifetime (13.2% [11.7-14.7]). It is important to note that fourteen of these women used the terms “induced miscarriage” or “therapeutic abortion” to describe induced abortion.

Table 4
Comparison of pregnancy outcomes in the COCON survey with the reference statistics (per 100 live births)

Our results show that varying the terms used to describe induced abortion helps to improve the specificity of the reports, by reducing the number of classification errors. The confusion in the vocabulary used by women on the subject of induced abortion was previously described in the qualitative GINE survey, in which the following statements were recorded : “I had mifepristone firstly because I wouldn’t even consider having an induced abortion” and “I didn’t have an induced abortion, I had a miscarriage after taking mifepristone” [2]. This confusion was also observed in the COCON survey. The most common error was to use the term “induced abortion” when talking about a different obstetrical event. Of 724 induced abortions reported at the beginning of the questionnaire, in response to a question using only the terms “induced abortion or termination”, 6.6% were not in fact induced abortions and were reclassified as therapeutic abortions or spontaneous miscarriages. The confusion between the terms “miscarriage” and “induced abortion” has been observed in other studies [1, 6] and is probably due in part to the discourse of the doctor, in which induced abortion, especially when performed under medical supervision, is sometimes presented as the equivalent in physiological terms of a miscarriage. Reclassification also allowed us to identify 13 induced abortions that were not mentioned in response to the “age at first induced abortion” question at the beginning of the questionnaire. It would therefore be useful for surveys, which usually use only the terms “induced abortion” or “termination” when referring to elective pregnancy termination, to include one or two additional questions to clarify the answers given to an initial general question on induced abortion, such as that used in COCON concerning age at first induced abortion. Women could then be asked if the pregnancy ended spontaneously or through medical intervention, and if the termination occurred at the woman’s own request or following a medical decision. In the latter case, it would be useful to ask the reasons for the termination, to allow for possible reclassification as a therapeutic abortion. This increased precision in reports would reduce the number of false-positives (reported as induced abortions but which are not), which seems to be non-negligible (6%) when only the term “induced abortion” is used in questions on this event.

This study also allowed us to compare the extent of underreporting of induced abortions depending on different wordings of questions in the same survey and in other recent French surveys. We found that despite the great care taken over question wording (in the sense of legitimizing induced abortion and diversifying the vocabulary employed), underestimation remained substantial in the COCON survey, at around 40%. This level of underreporting seems to represent an absolute minimum in studies on the general population, to judge from the consistency of the phenomenon over time and space regardless of the mode of questioning and the survey aims. The stability of this underreporting suggests we need to think again about its causes and to consider possibilities other than the difficulty of talking about sensitive events. The sources of this underestimation may be multiple.

It could come in part from the sampling method used in COCON, which, as in all surveys on the general population, does not allow perfect coverage of the target population, and underestimates the marginalized and the most disadvantaged sections of the population, where recourse to abortion may be more frequent. It must also be noted that 26% of eligible women refused to participate in the survey. Even though sampling distortions were taken into account by adjusting the sample to the distribution of the main socio-demographic characteristics of the French female population, this does not guarantee a total absence of bias since it assumes that the women “not interviewed” would have responded in the same way as the average of the women with the same profile who did respond. This is not at all certain in a study on induced abortion.

The magnitude of the underreporting of induced abortion revealed in the survey cannot however be due solely to a problem of target population coverage. Also involved is the difficulty of talking about intimate events that carry low social approval. The study by Jagannathan in the United States showed that women are less likely to report an induced abortion if they have a negative attitude towards abortion [12]. Likewise, the lower rate of underreporting in Romania, where abortion was encouraged by non-restrictive policies and remains common, suggests that social acceptability influences reporting.

The context in which the induced abortion occurs, how the event is understood and experienced, may also affect women’s propensity to talk about it. A comparison of the socio-demographic profiles of women who had abortions in the COCON survey with national data derived from the compulsory reporting forms shows that very young women (< 20 years), older women (> 40 years) and women living alone are underrepresented. Other surveys have already shown that the rate of underreporting of induced abortions is higher in women aged over 40 and those living alone [1, 6, 12]. Abortion often acquires a special significance for these women — for the youngest because of their unacknowledged sexuality, for the oldest because of the uncompromising nature of the decision to terminate — which may make it hard for them to talk about it. We also have to consider the effect of the emotional context for the abortion, possibly linked to a relational episode that occupies an uncertain place in the women’s sexual history. Previous studies have shown that women do not always report all of their sexual partners and that in particular they do not mention those who counted for little in their lives [15]. The question then arises of whether the same could be true for the events connected with these relationships.

However, if the social stigma attached to induced abortion was the only explanation, we would expect the emergence of greater social acceptance for this event to bring about a reduction in underreporting over time. Similarly, we would expect the inclusion in the COCON survey of a question whose wording legitimized induced abortion, to have improved the reporting rate, but this was not the case. Conversely, the fact that our study was carried out by a public medical research institute (INSERM) may have encouraged women to give responses that were consistent with the contraceptive norm, thus counterbalancing the legitimization effect aimed for in the questionnaire.

In this respect it should be emphasized that the underreporting of health-related events is not specific to induced abortion, as has been shown by several studies on spontaneous miscarriage [9, 14, 24]. Wilcox and Horney showed that only 75% of spontaneous miscarriages initially recorded for the purposes of follow-up were reported when the women were questioned retrospectively after the observation period [24]. The extent of underreporting was inversely proportional to the duration of pregnancy, with the earliest miscarriages being the most frequently omitted. Likewise, a Norwegian survey based on the study of consistency between reports from women and medical data about them from registers showed that 26.5% of spontaneous miscarriages mentioned in the registers went unreported by the women, and that the likelihood of reporting a miscarriage decreased with time [14]. This screening-out phenomenon is not limited to episodes associated with reproductive life but is observed for health-related events in general, be they acute or chronic, serious or benign. In a study conducted in 1992 on the GAZEL cohort (comprising employees of the French gas and electricity company, EDF-GDF), Metzger et al. showed that, for chronic health problems causing an absence from work in the year preceding the survey, the proportion reported by subjects varied between 8.9% and 100% depending on the type of condition (psychiatric or nervous disorders 8.9%, phlebitis 41%, cancers 71.7%, ulcers 75%, diabetes 95%) [16]. The underreporting of episodes of ill health depended on the individual characteristics of subjects, the characteristics of the condition (severity and disability) and the characteristics of the survey (recall bias, precision of question wording). Similarly, periods of hospitalization are often omitted in surveys : the study by Norrish et al. in New Zealand, showed that 32% of first admissions and 55% of readmissions in the four years prior to the survey were not reported by the patients [17]. These considerations lead us to reconsider the question of the underreporting of induced abortion, not only in terms of the underreporting of sensitive events that represent a failure of contraceptive practice, but also in terms of the screening-out of medical events in general, as if reporting a health event would contribute to the construction of a “sick person” identity, which has low social prestige.

To test these hypotheses, we would need to distinguish between the non-reporting of induced abortions originating in the screening-out of health-related events and that deriving from the stigmatization of induced abortion. In this perspective, and as a complement to the questions directly concerned with abortion, it would be valuable to collect information on health problems in general, comparing for example the reported number of hospitalizations in a year with those from official hospital data. To measure the effect of social stigmatization on the underreporting of induced abortions, abortion rates for the women surveyed could usefully be compared with those for their close friends and relatives, these latter rates being estimated indirectly by asking the women in the survey if they know of friends and relatives who have had an induced abortion (their three closest confidants for example). This method has proved valuable in several contexts, notably for the reporting of homosexuality in France in the survey on sexual behaviour (ACSF) [5] and for estimating the rate of induced abortion where recourse to abortion is illegal [18].

If the woman said that she had not talked about this pregnancy, she was asked to give its date and outcome. Furthermore, a few of the possible answers in the “Decision concerning last unplanned pregnancy” and “Access to health care for last induced abortion” modules supplied additional information in the cases where it was hard to decide between induced abortion and some other pregnancy outcome (induced miscarriage or therapeutic abortion). This concerned the answer “It was not me that made the decision (medical decision)” to the question “Was the decision to terminate the pregnancy taken immediately after a short or a long hesitation ?” and the answer “Abortion clinic” to the question “Can you tell me all the people you contacted when making your request for induced abortion or termination ?”