Contraceptive confidence and timing of first birth in Moldova: an event history analysis of retrospective data

Objectives To test the contraceptive confidence hypothesis in a modern context. The hypothesis is that women using effective or modern contraceptive methods have increased contraceptive confidence and hence a shorter interval between marriage and first birth than users of ineffective or traditional methods. We extend the hypothesis to incorporate the role of abortion, arguing that it acts as a substitute for contraception in the study context. Setting Moldova, a country in South-East Europe. Moldova exhibits high use of traditional contraceptive methods and abortion compared with other European countries. Participants Data are from a secondary analysis of the 2005 Moldovan Demographic and Health Survey, a nationally representative sample survey. 5377 unmarried women were selected. Primary and secondary outcome measures The outcome measure was the interval between marriage and first birth. This was modelled using a piecewise-constant hazard regression, with abortion and contraceptive method types as primary variables along with relevant sociodemographic controls. Results Women with high contraceptive confidence (modern method users) have a higher cumulative hazard of first birth 36 months following marriage (0.88 (0.87 to 0.89)) compared with women with low contraceptive confidence (traditional method users, cumulative hazard: 0.85 (0.84 to 0.85)). This is consistent with the contraceptive confidence hypothesis. There is a higher cumulative hazard of first birth among women with low (0.80 (0.79 to 0.80)) and moderate abortion propensities (0.76 (0.75 to 0.77)) than women with no abortion propensity (0.73 (0.72 to 0.74)) 24 months after marriage. Conclusions Effective contraceptive use tends to increase contraceptive confidence and is associated with a shorter interval between marriage and first birth. Increased use of abortion also tends to increase contraceptive confidence and shorten birth duration, although this effect is non-linear—women with a very high use of abortion tend to have lengthy intervals between marriage and first birth.


Introduction
Over the last two decades, most countries in Eastern Europe have experienced an unprecedented decline in fertility rates either at or below 1.3 children per woman 1,2 .
Economic uncertainty and high male out-migration partly explain the stagnant low fertility trends, although recent data show gradual recovery of fertility rates in some countries 1,3 .
Many women in Eastern Europe tend to control their fertility by using traditional contraceptive methods or induced abortions since modern method access is limited 4,5,6 .
This research focusses on Moldova where abortions are widely practised and often accepted as a birth control method.
The dynamics of contraceptive use including discontinuation rates, switching and method efficacy is widely acknowledged in demographic research 7,8,9,10 . However, the confidence which women have in their contraceptive method and the effect it has on fertility behaviour is under-researched. Contraceptive confidence is an hypothesis which explains timing of childbearing resulting from the perceived efficacy of contraceptive methods 11, 12, 13 . Theoretically, women who use less effective contraceptive methods (traditional methods) have low contraceptive confidence, since their method is likely to fail.
These women tend to space their fertility as a means to limit their intended family size 12 . In contrast, women who use effective (modern) contraceptives have a high degree of confidence that these methods will not fail. This has prompted women to compress their fertility into shorter periods 14,15 .
While previous studies have addressed second and later birth intervals, the demographic landscape of Europe has undergone unprecedented changes in recent decades driven mostly by changes in the relationship between partnership formation-particularly marriage-and childbearing 1,16 . These trends are gradually emerging in Moldova signalling the features of a second demographic transition 17 exemplified mostly in terms of low fertility  From the original MDHS sample of 7,440 women, 1,884 women were excluded since they were never married and 74% of these reported having never had sex. In addition, 179 women who had premarital births (2.4%) were excluded since the terminal event (first birth)

Method
The analysis uses a piecewise-constant hazard model. The dependent variable is the timing of first birth (terminal event) since first marriage (start event), recorded in months and expressed as ) (t y i , a binary random variable for each time piece following marriage, where; if woman i has a birth at time t, and 0 ) ( = t y i if woman i does not experience birth at t. The hazard of a first birth is defined as  Since the final model includes many interactions, the interpretation of the coefficient directly is extremely difficult. Therefore, we use the model to generate survival curves and cumulative hazards, which are presented for interpretation.

Explanatory variables
The main interest in the analysis of first birth interval is the degree of contraceptive confidence. As noted by Ní Bhrolcháin 12 , the perfect measure of contraceptive confidence would include information on contraceptive tastes and preferences collected contemporaneously with use. Ní Bhrolcháin 12, 14, 15 argues that in the absence of this information the best available proxy is the most recent contraceptive method. We note that that women may have changed their contraceptive method since their first birth, and hence our estimated contraceptive confidence may not necessarily correspond to the method used preceding the first birth. While the MDHS does include data on contemporary contraceptive use in the contraceptive calendar, this data pertains to the 5 years prior to the survey. Using these data is not considered feasible since a) there is only a small number of first births in that interval (fewer than 140) and b) the recency of the births would severely constrain our ability to make inference particularly for older marriage cohorts. About 57% of sexually active women in the MDHS have reported not switching their contraceptive method within the past 5 years. This is an important observation which validates the assumption that women in Moldova are unlikely to switch their contraceptive method. This analysis defines low contraceptive confidence for women who reported using a traditional method (22% of women use either withdrawal or periodic abstinence), moderate contraceptive confidence for those using a modern reversible method (e.g. pill, condom, IUD, constituting 36%) and high contraceptive confidence for women using a permanent method (5%) either female or male. About 37% of women in the analysis sample have reported not using any method: contraceptive confidence for these women cannot be observed. We retain these women in the analysis however, since their abortion history is still important in a context where abortion is normative fertility control behaviour. We include two controls relevant to contraceptive behaviour: the month and year of first method use and another variable measuring the previous method discontinued.
To capture the latent effect of abortion propensity, the analysis uses abortion history as a proxy measure. Unfortunately, the MDHS has not collected any data on abortion attitudes. We therefore use the proportion of pregnancies a woman has terminated. A simple count is inadequate since older women have greater exposure to multiple abortions, which may introduce bias. Using the proportion of pregnancies aborted overcomes this problem. Other than recall problems inherent in cross-sectional surveys, any deliberate under-reporting of abortion in post-Socialist countries is very low 20,27 . Contraception and abortion are often seen complementary in the Moldovan context-women report that the use of ineffective methods (such as withdrawal) combined with frequent recourse to abortion is a normative fertility control technique especially for traditional method users. An interaction between contraceptive method and abortion propensity is used to test the differential effect of abortion on different levels of contraceptive confidence.
Another key predictor variable is marriage cohort intended to capture the changes in first birth rate which is often determined by economic circumstances especially the there should be some caution when interpreting results for the oldest marriage cohort since there will be some left censoring: this marriage cohort is specified covering a wider range than others to ensure sufficient sample size. The model controls for other effects which could potentially influence the decision to have a first birth, the ability of women to conceive and socio-demographic characteristics. These include: age at marriage, level of education of women, geographical region and place of residence. As with the key explanatory variables some of these are proxy variables limited to information available at survey. For example, the duration of the first marriage is used to estimate whether the woman was in a continuous marital union prior to first birth and whether union dissolution or separation occurred before the first birth. Other control variables were considered in the model as they were thought to be relevant a priori (e.g. ethnicity, wealth index), but were found not to significantly improve the model fit. Statistical significance was assessed by the use of the Likelihood Ratio (LR) test with significance at the 5% level. The model was estimated in SPSS 19.0.

Results
The regression results adjusting for relevant confounders and control variables are presented for three selected effects (i) marriage cohorts, (ii) contraceptive confidence and (iii) abortion propensity. Due the interaction terms and time dependency specified in the model, it is difficult to interpret coefficient directly. We therefore use this model to generate estimated survival curves and cumulative hazards, and report the cumulative hazard of first birth at 12, 24 and 36 months after marriage as a summary statistic in Table 1 as well as cumulative survival curves for each main variable examined.

Marriage cohorts
The adjusted hazard rate of a first birth for each duration since marriage is estimated for different marriage cohorts. The results are shown in the form of survival plots (Figure 1), truncated at 36 months for visual clarity. The survival plot indicates the proportion of women yet to have first birth at month t following marriage. We also report the cumulative hazard of first birth at 12, 24 and 36 months after marriage as a summary statistic in Table 1 a).
<< Table 1  overlaps with the most recent cohort after 24 months which suggests the propensity for early transition to motherhood among recently married women. That said, the overall probability of having a birth remains relatively constant-for instance 3 years following marriage the later cohorts have attained the same proportion having had a birth as the pre-Socialist marriage cohorts. This is largely due to recuperation effect 2 to 3 years following marriage, suggesting that although the interval between marriage and first birth is longer, the probability of giving a birth does not vary across cohorts. This is also reflected in the cumulative hazard, with the hazard among the preindependence cohorts at 41%, 75% and 86% for 12, 24 and 36 months respectively.
However, there is a considerable fall in the cumulative hazard for the 1995-99 and 2000 marriage cohorts, indicating increasing delay of first birth following the collapse of Socialism, but overall Moldovan women have a consistently high probability of becoming mothers.

Contraceptive confidence
The estimated survival curve for each level of contraceptive confidence is presented in Figure 2. Cumulative hazards are presented in Table 1 b). Due to the interaction between contraceptive confidence and abortion propensity, these estimated survival plots are generated where the categories of abortion propensity are set to their sample proportions.
All other covariates are held constant, producing net effects controlling for selected characteristics controlling for marriage cohort effects and socio-economic characteristics. <<Figure 2 about here>> Among women with a measurable contraceptive level (i.e. where a contraceptive method is recorded at survey), the survival curve for high contraceptive confidence is the highest, indicating the slowest transition to first birth in this group. Compared to women with a low contraceptive confidence, the first birth rate is higher for women with a moderate contraceptive confidence. The survival curve for high contraceptive confidence is comparable to those of the low confidence group until 24 months following marriage, when there is a rapid fall in the proportion of women yet to have first births. This indicates that, in general, low contraceptive confidence is associated with a low hazard of a first birth and -10 -hence longer duration between marriage and first birth. On the other hand, an increase in contraceptive confidence is associated with an increased hazard of a first birth, which clearly suggests rapid transition to motherhood among women with high confidence.

Abortion
The estimated survival curve of first birth for women with low contraceptive confidence is presented in Figure 3, which examines the association between low contraceptive confidence and abortion propensity. In general, the proportion of women yet to have a first birth is high for women with no abortion propensity, and the survival curves are lower for women with low and moderate abortion propensity. Table 1 c) presents the estimated cumulative hazard of first birth. Broadly, we see that the probability of having a first birth is low for women with no abortion propensity. However, among abortion users, the cumulative hazard of first birth is higher at 12, 24 and 36 months following marriage. This suggests that a higher propensity to use abortion reduces the interval between marriage and first birth. The survival curve for women with a high abortion propensity is roughly comparable or slightly lower than women with no abortion propensity. <<Figure 3 about here>>

Key points
• Contraceptive confidence influences the duration between marriage and first birth in Moldova • There is a distinct effect of abortion on contraceptive confidence: the availability of abortion tends to increase women's contraceptive confidence • The effect of macro-economic shocks and social transitions are evident on marriage cohort specific first birth rates

Strengths and limitations of this study
• Study uses a nationally representative survey • Use of regression analysis disentangles net effects of related contraceptive and abortion behaviour • Use of retrospective data means reliance on proxy measures.

Introduction
Over the last two decades, many countries in Eastern Europe have experienced an unprecedented decline in fertility rates either at or below 1.3 children per woman 1, 2 .
Economic uncertainty and high male out-migration partly explain the stagnant low fertility trends, although recent data show gradual recovery of fertility rates in some countries 1, 3 .
Many women in Moldova tend to control their fertility by using traditional contraceptive methods or induced abortions since modern method access is limited 4,5,6 . This research focusses on Moldova where abortions are widely practised and often accepted as a birth control method.
The dynamics of contraceptive use including discontinuation rates, switching and method efficacy is widely acknowledged in demographic research 7,8,9, 10, c . However, the confidence which women have in their contraceptive method and the effect it has on fertility behaviour is under-researched. Contraceptive confidence is an hypothesis which explains timing of childbearing resulting from the perceived efficacy of contraceptive methods, but there is little modern literature 11 and much work examines older demographic data 12, 13 . Theoretically, women who use less effective contraceptive methods (traditional methods) have low contraceptive confidence, since their method is likely to fail. These women tend to space their fertility as a means to limit their intended family size 12 . In contrast, women who use effective (modern) contraceptives have a high degree of confidence that these methods will not fail. This has prompted women to compress their fertility into shorter periods 14, 15 .
While previous studies have addressed second and later birth intervals, the demographic landscape of Europe has undergone unprecedented changes in recent decades driven mostly by changes in the relationship between partnership formation-particularly marriage-and childbearing 1,16 . These trends are gradually emerging in Moldova signalling We note that the pattern of union formation is an exceptionally complex demographic process 11 . As well as the control variables we are able to include, there will typically be significant variation in behaviour that are important but not captured by the type of representative sample survey we employ. Therefore, while we are able to describe part of the effects on first birth, this analysis should not be interpreted as a complete picture.
In the Moldova, traditional methods are still widely used: about 26% of the contraceptive methods used in Moldova are traditional 20, 21 . This is considerably higher than observed even in other former-Soviet countries (Latvia 8.7%, Hungary 9.0% and Bulgaria 15.7%) 4 . Moldova therefore lends itself to examining the differential effects of contraceptive confidence on reproductive behaviour. Another characteristic of fertility control behaviour in Moldova is the widespread use of abortion-46% of ever-sexually active women reported having had at least one abortion and about 40% of these women have had two or more abortions 21 . Widespread use of traditional contraceptives and method failure are associated   From the original MDHS sample of 7,440 women, 1,884 women were excluded since they were never married and 74% of these reported having never had sex. In addition, 179 women who had premarital births (2.4%) were excluded since the terminal event (first birth) preceded the start event (marriage). The final selected sample considers 5,377 married women. About 15% of births occurred within 9 months of marriage-indicative of premarital conception. The MDHS also include detailed information of abortion histories including the number and timing of each abortion.

Method
The analysis uses a piecewise-constant hazard model. The dependent variable is the timing of first birth (terminal event) since first marriage (start event), recorded in months and expressed as ) (t y i , a binary random variable for each time piece following marriage, where; if woman i has a birth at time t, and 0 ) ( = t y i if woman i does not experience birth In equation 1, ( )

Explanatory variables
The main interest in the analysis of first birth interval is the degree of contraceptive confidence. As noted by Ní Bhrolcháin 12 , the perfect measure of contraceptive confidence would include information on contraceptive tastes and preferences collected contemporaneously with use. Ní Bhrolcháin 12, 14, 15 argues that in the absence of this information the best available proxy is the most recent contraceptive method. We note that that women may have changed their contraceptive method since their first birth, and hence our estimated contraceptive confidence may not necessarily correspond to the method used preceding the first birth. While the MDHS does include data on contemporary contraceptive use in the contraceptive calendar, this data pertains to the 5 years prior to the survey. Using these data is not considered feasible since a) there is only a small number of first births in that interval (fewer than 140) and b) the recency of the births would severely constrain our ability to make inference particularly for older marriage cohorts. About 57% of sexually active women in the MDHS have reported not switching their contraceptive method within the past 5 years. This is an important observation which validates the assumption that women in Moldova are unlikely to switch their contraceptive method.
This analysis defines low contraceptive confidence for women who reported using a traditional method (22% of women use either withdrawal or periodic abstinence), moderate contraceptive confidence for those using a modern reversible method (e.g. pill, condom, IUD, constituting 36%) and high contraceptive confidence for women using a permanent method (5%) either female or male. About 37% of women in the analysis sample have reported not using any method: contraceptive confidence for these women cannot be observed. We retain these women in the analysis however, since their abortion history is still important in a context where abortion is normative fertility control behaviour. We include two controls relevant to contraceptive behaviour: the month and year of first method use and another variable measuring the previous method discontinued.
To capture the latent effect of abortion propensity, the analysis uses abortion history as a proxy measure. Unfortunately, the MDHS has not collected any data on abortion attitudes. We therefore use the proportion of pregnancies a woman has terminated. A simple count is inadequate since older women have greater exposure to multiple abortions, which may introduce bias. Using the proportion of pregnancies aborted overcomes this problem. Other than recall problems inherent in cross-sectional surveys, any deliberate under-reporting of abortion in post-Socialist countries is very low 22, 28 . Contraception and abortion are often seen complementary in the Moldovan context-women report that the use of ineffective methods (such as withdrawal) combined with frequent recourse to abortion is a normative fertility control technique especially for traditional method users. An interaction between contraceptive method and abortion propensity is used to test the differential effect of abortion on different levels of contraceptive confidence.
Another key predictor variable is marriage cohort intended to capture the changes in first birth rate which is often determined by economic circumstances especially the availability of housing 23, 27, 29 . The age range of women in the dataset (15-49) means that there should be some caution when interpreting results for the oldest marriage cohort since there will be some left censoring: this marriage cohort is specified covering a wider range than others to ensure sufficient sample size. The model controls for other effects which could potentially influence the decision to have a first birth, the ability of women to conceive and socio-demographic characteristics. These include: age at marriage, level of education of

Results
The regression results adjusting for relevant confounders and control variables are presented for three selected effects (i) marriage cohorts, (ii) contraceptive confidence and (iii) abortion propensity. The full model is presented in Table 1A, which is made available separately. Due the interaction terms and time dependency specified in the model, it is difficult to interpret coefficient directly, in particular the assessment of statistical significance of overall probabilities. We therefore use this model to generate estimated survival curves and cumulative hazards, and report the cumulative hazard of first birth at 12, 24 and 36 months after marriage as a summary statistic in Table 1 as well as cumulative survival curves for each main variable examined. In the tables, to allow the reader to assess significant effects, we present confidence intervals adjusted for pairwise comparisons at the 5% level: the non-overlap of these intervals can be interpreted as a difference which is significant at the 5% level.

Marriage cohorts
The adjusted hazard rate of a first birth for each duration since marriage is estimated for different marriage cohorts. The results are shown in the form of survival plots (Figure 1), truncated at 36 months for visual clarity. The survival plot indicates the proportion of women yet to have first birth at month t following marriage. We also report the cumulative hazard of first birth at 12, 24 and 36 months after marriage as a summary statistic in Table 1 a).
<< Table 1  Socialist marriage cohorts. This is largely due to recuperation effect 2 to 3 years following  This is also reflected in the cumulative hazard, with the hazard among the preindependence cohorts at 41%, 75% and 86% for 12, 24 and 36 months respectively.
However, there is a considerable fall in the cumulative hazard for the 1995-99 and 2000 marriage cohorts, indicating increasing delay of first birth following the collapse of Socialism, but overall Moldovan women have a consistently high probability of becoming mothers.

Contraceptive confidence
The estimated survival curve for each level of contraceptive confidence is presented in Figure 2. Cumulative hazards are presented in Table 1 b). Due to the interaction between contraceptive confidence and abortion propensity, these estimated survival plots are generated where the categories of abortion propensity are set to their sample proportions.
All other covariates are held constant, producing net effects controlling for selected characteristics controlling for marriage cohort effects and socio-economic characteristics.

<<Figure 2 about here>>
Among women with a measurable contraceptive level (i.e. where a contraceptive method is recorded at survey), the survival curve for high contraceptive confidence is the highest, indicating the slowest transition to first birth in this group. Compared to women with a low contraceptive confidence, the first birth rate is higher for women with a moderate contraceptive confidence. The survival curve for high contraceptive confidence is comparable to those of the low confidence group until 24 months following marriage (indeed there is no detectable statistically significant difference at this point), when there is a rapid fall in the proportion of women yet to have first births. This indicates that, in general, low contraceptive confidence is associated with a low hazard of a first birth and hence longer duration between marriage and first birth. On the other hand, an increase in contraceptive confidence is associated with an increased hazard of a first birth, which clearly suggests rapid transition to motherhood among women with high confidence.

Abortion
The estimated survival curve of first birth for women with low contraceptive confidence is presented in Figure 3, which examines the association between low contraceptive confidence and abortion propensity. In general, the proportion of women yet to have a first birth is high for women with no abortion propensity, and the survival curves are lower for women with low and moderate abortion propensity. Table 1 c) presents the estimated cumulative hazard of first birth. Broadly, we see that the probability of having a first birth is low for women with no abortion propensity. However, , the cumulative hazard of first birth is significantly higher at 12, 24 and 36 months among low and 12 and 24 months moderate abortion users following marriage. This suggests that overall women who were prepared to use abortion-at least partially have a shorter interval between marriage and first birth. The survival curve for women with a high abortion propensity is roughly comparable or slightly lower than women with no abortion propensity. We cannot detect an effect for high abortion prevalence. Indeed there is some evidence of an attenuation in the higher cumulative hazard of first birth at higher abortion levels: at 12, 24 and 26 months the cumulative hazard is lower for moderate and high abortion users than women with a low abortion propensity.

Conclusion
This paper examined the impact of contraceptive confidence on the shifts in timing of first birth in a low fertility regime with high abortion rates. The analysis yielded three key findings. First, there is evidence of contraceptive confidence effect on the timing of first birth: women with low contraceptive confidence tend to delay their first birth, while women with high contraceptive confidence progress more rapidly to motherhood. The results supported the hypothesis that women using effective methods have increased contraceptive confidence and have relatively shorter interval between marriage and first birth than users of ineffective methods. This result has wide ranging implications in the low fertility context of Moldova where modern methods are not widely available and many women rely on traditional methods for fertility control. Second, overall use of abortion results in shorter interval between marriage and first birth particularly for women with a low contraceptive confidence. We do note however that this effect is non-linear: increasing propensity to use abortion (for example high compared to low propensity) will tend to depress overall fertility behaviour. Abortion appears to be an effective substitute for women with low contraceptive confidence, suggesting that voluntary abortion tend to potentially outweigh a traditional method failure. An efficient strategy to reduce increasing abortion rates, therefore, is to increase access to modern methods to young couples in Moldova. Third, the study provides evidence of an increase in the duration between marriage and first birth for more recent marriage cohorts although motherhood is still common among Moldovan women. This development is consistent with the increasing trend in fertility postponement behaviour as well as increasingly complex co-relationships between fertility and marriage in the Moldovan    Contributorship: Author 1 contributed to paper design and conceptualisation, data analysis and drafting the manuscript. Authors 2 and 3 contributed to paper design and conceptualisation and drafting the manuscript.

Competing interests: None
Data Sharing Statement: There are no additional data beyond those stated in the article, which are freely available from OCR Macro upon request.        1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  • Contraceptive confidence influences the duration between marriage and first birth in 5 Moldova 6 7 • There is a distinct effect of abortion on contraceptive confidence: the availability of 8 abortion tends to increase women's contraceptive confidence 9 10  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  Over the last two decades, most many countries in Eastern Europe have experienced an 3 unprecedented decline in fertility rates either at or below 1.3 children per woman 1, 2 . 4 Economic uncertainty and high male out-migration partly explain the stagnant low fertility 5 trends, although recent data show gradual recovery of fertility rates in some countries 1, 3 . 6 Many women in Eastern EuropeMoldova tend to control their fertility by using traditional 7 contraceptive methods or induced abortions since modern method access is limited 4,5,6 . 8 This research focusses on Moldova where abortions are widely practised and often accepted 9 as a birth control method. 10 The dynamics of contraceptive use including discontinuation rates, switching and 11 method efficacy is widely acknowledged in demographic research 7,8,9, 10, C . However, the 12 confidence which women have in their contraceptive method and the effect it has on 13 fertility behaviour is under-researched. Contraceptive confidence is an hypothesis which 14 explains timing of childbearing resulting from the perceived efficacy of contraceptive 15 methods 11, 12, 13 . Theoretically, women who use less effective contraceptive methods 16 (traditional methods) have low contraceptive confidence, since their method is likely to fail. 17 These women tend to space their fertility as a means to limit their intended family size 12 . In 18 contrast, women who use effective (modern) contraceptives have a high degree of 19 confidence that these methods will not fail. This has prompted women to compress their  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  Therefore, any analysis exploring fertility 8 behaviour should account for marriage cohort as an important control variable, albeit not 9 one that can offer a complete explanation of observed trends. 10 We note that the pattern of union formation is an exceptionally complex 11 demographic process 11 . As well as the control variables we are able to include, there will 12 typically be significant variation in behaviour that are important but not captured by the 13 type of representative sample survey we employ. Therefore, while we are able to describe 14 part of the effects on first birth, this analysis should not be interpreted as a complete 15

Method 18
The analysis uses a piecewise-constant hazard model. The dependent variable is the timing 19 of first birth (terminal event) since first marriage (start event), recorded in months and 20 expressed as ) (t y i , a binary random variable for each time piece following marriage, where; 21 if woman i has a birth at time t, and 0 ) ( = t y i if woman i does not experience birth 22 at t. The hazard of a first birth is defined as Eq. (1) 5 In equation 1, ( ) contraceptive confidence and propensity to use abortion is specified in addition to the main 20

effects.
Since the final model includes many interactions, the interpretation of the 21 coefficient directly is extremely difficult. Therefore, we use the model to generate survival 22 curves and cumulative hazards, which are presented for interpretation. 23

Explanatory variables 1
The main interest in the analysis of first birth interval is the degree of contraceptive 2 confidence. As noted by Ní Bhrolcháin 12 , the perfect measure of contraceptive confidence 3 would include information on contraceptive tastes and preferences collected 4 contemporaneously with use. Ní Bhrolcháin 12, 14, 15 argues that in the absence of this 5 information the best available proxy is the most recent contraceptive method. We note that 6 that women may have changed their contraceptive method since their first birth, and hence 7 our estimated contraceptive confidence may not necessarily correspond to the method used 8 preceding the first birth. While the MDHS does include data on contemporary contraceptive 9 use in the contraceptive calendar, this data pertains to the 5 years prior to the survey. Using 10 these data is not considered feasible since a) there is only a small number of first births in 11 that interval (fewer than 140) and b) the recency of the births would severely constrain our 12 ability to make inference particularly for older marriage cohorts. About 57% of sexually 13 active women in the MDHS have reported not switching their contraceptive method within 14 the past 5 years. This is an important observation which validates the assumption that 15 women in Moldova are unlikely to switch their contraceptive method. 16 17 This analysis defines low contraceptive confidence for women who reported using a 18 traditional method (22% of women use either withdrawal or periodic abstinence), moderate 19 contraceptive confidence for those using a modern reversible method (e.g. pill, condom, 20 IUD, constituting 36%) and high contraceptive confidence for women using a permanent 21 method (5%) either female or male. About 37% of women in the analysis sample have 22 reported not using any method: contraceptive confidence for these women cannot be 23 observed. We retain these women in the analysis however, since their abortion history is still 24 important in a context where abortion is normative fertility control behaviour. We include 25   1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59

Results 8
The regression results adjusting for relevant confounders and control variables are 9 presented for three selected effects (i) marriage cohorts, (ii) contraceptive confidence and 10 (iii) abortion propensity. The full model is presented in Table 1A, which is made available 11 separately. Due the interaction terms and time dependency specified in the model, it is 12 difficult to interpret coefficient directly, in particular the assessment of statistical 13 significance of overall probabilities . We therefore use this model to generate estimated 14 survival curves and cumulative hazards, and report the cumulative hazard of first birth at 12, 15 24 and 36 months after marriage as a summary statistic in Table 1 as well as cumulative 16 survival curves for each main variable examined. In the tables, to allow the reader to assess 17 significant effects, we present confidence intervals adjusted for pairwise comparisons at the 18 5% level: the non-overlap of these intervals can be interpreted as a difference which is 19 significant at the 5% level.  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  women yet to have first birth at month t following marriage. We also report the cumulative 1 hazard of first birth at 12, 24 and 36 months after marriage as a summary statistic in Table 1  Socialist marriage cohorts. This is largely due to recuperation effect 2 to 3 years following 19 marriage, suggesting that although the interval between marriage and first birth is longer, 20 the probability of giving a birth does not vary across cohorts. 21 This is also reflected in the cumulative hazard, with the hazard among the pre-22 independence cohorts at 41%, 75% and 86% for 12, 24 and 36 months respectively.  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59

Abortion 4
The estimated survival curve of first birth for women with low contraceptive confidence is 5 presented in Figure 3, which examines the association between low contraceptive 6 confidence and abortion propensity. In general, the proportion of women yet to have a first 7 birth is high for women with no abortion propensity, and the survival curves are lower for 8 women with low and moderate abortion propensity. Table 1 c) presents the estimated 9 cumulative hazard of first birth. Broadly, we see that the probability of having a first birth is 10 low for women with no abortion propensity. However, among abortion users, the 11 cumulative hazard of first birth is significantly higher at 12, 24 and 36 months among low 12 and 12 and 24 months moderate abortion users following marriage. This suggests that a 13 higher propensityoverall women who were prepared to use abortion-at least partially to use 14 abortion reduceshave a shorter interval between the interval between marriage and first 15 birth. The survival curve for women with a high abortion propensity is roughly comparable or 16 slightly lower than women with no abortion propensity. We cannot detect an effect for high 17 abortion prevalence. Indeed there is some evidence of an attenuation in the higher 18 cumulative hazard of first birth at higher abortion levels: at 12, 24 and 26 months the 19 cumulative hazard is lower for moderate and high abortion users than women with a low 20 abortion propensity.  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59

• Contraceptive confidence influences the duration between marriage and first birth in Moldova
• There is a distinct effect of abortion on contraceptive confidence: the availability of abortion tends to increase women's contraceptive confidence • The effect of macro-economic shocks and social transitions are evident on marriage cohort specific first birth rates

Strengths and limitations of this study
• Study uses a nationally representative survey • Use of regression analysis disentangles net effects of related contraceptive and abortion behaviour • Use of retrospective data necessitates reliance on proxy measures. Over the last two decades, many countries in Eastern Europe have experienced an 3 unprecedented decline in fertility with a Total Fertility Rate at or below 1.3 children per 4 woman 1, 2 . Economic uncertainty and high male out-migration partly explain the stagnant 5 low fertility trends, although recent data show gradual recovery of fertility rates in some 6 countries 1, 3 . Many women in Moldova tend to control their fertility by using traditional 7 contraceptive methods or induced abortions since modern method access is limited 4,5,6 . 8 This research focusses on Moldova where abortions are widely practised and often accepted 9 as a birth control method. 10 The dynamics of contraceptive use including discontinuation rates, switching and 11 method efficacy is widely acknowledged in demographic research 7,8,9,10, . However, the 12 confidence which women have in their contraceptive method and the effect it has on 13 fertility behaviour is under-researched. Contraceptive confidence is an hypothesis which 14 explains timing of childbearing resulting from the perceived efficacy of contraceptive 15 methods, but there is little modern literature 11 and much work examines older demographic 16 data 12, 13 . Theoretically, women who use less effective contraceptive methods (traditional 17 methods) have low contraceptive confidence, since their method is likely to fail. These 18 women tend to space their fertility as a means to limit their intended family size 12 . In 19 contrast, women who use effective (modern) contraceptives have a high degree of 20 confidence that these methods will not fail. This has prompted women to compress their 21 fertility into shorter periods 14,15 . 22 While previous studies have addressed second and later birth intervals, the 23 demographic landscape of Europe has undergone unprecedented changes in recent decades 24 driven mostly by changes in the relationship between partnership formation-particularly 25 marriage-and childbearing 1,16 . These trends are gradually emerging in Moldova signalling 26  have been a number of other explanations for changing fertility across Eastern Europe (for 7 example, more orthodox economic factors), and the cause is still debated among 8 demographers and dependent on context 19,20 . Therefore, any analysis exploring fertility 9 behaviour should account for marriage cohort as an important control variable, albeit not 10 one that can offer a complete explanation of observed trends. 11 We note that the pattern of union formation is an exceptionally complex 12 demographic process 11 . As well as the control variables we are able to include, there will 13 typically be significant variation in behaviour that are important but not captured by the 14 type of representative sample survey we employ. Therefore, while we are able to describe 15 part of the effects on first birth, this analysis should not be interpreted as a complete 16

picture. 17
In the Moldova, traditional methods are still widely used: about 26% of the 18 contraceptive methods used in Moldova are traditional 20, 21 . This is considerably higher than 19 observed even in other former-Socialist countries (Latvia 8.7%, Hungary 9.0% and Bulgaria 20 15.7%) 4 . Moldova therefore lends itself to examining the differential effects of contraceptive 21 confidence on reproductive behaviour. Another characteristic of fertility control behaviour in 22 Moldova is the widespread use of abortion-46% of ever-sexually active women reported 23 having had at least one abortion and about 40% of these women have had two or more 24 abortions 21 . Widespread use of traditional contraceptives and method failure are associated 25   1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59   From the original MDHS sample of 7,440 women, 1,884 women were excluded since 11 they were never married and 74% of these reported having never had sex. In addition, 179 12 women who had premarital births (2.4%) were excluded since the terminal event (first birth) 13

Explanatory variables 4
The main interest in the analysis of first birth interval is the degree of contraceptive 5 confidence. As noted by Ní Bhrolcháin 12 , the perfect measure of contraceptive confidence 6 would include information on contraceptive tastes and preferences collected 7 contemporaneously with use. Ní Bhrolcháin 12, 14, 15 argues that in the absence of this 8 information the best available proxy is the most recent contraceptive method. We note that 9 that women may have changed their contraceptive method since their first birth, and hence 10 our estimated contraceptive confidence may not necessarily correspond to the method used 11 preceding the first birth. While the MDHS does include data on current contraceptive use in 12 the contraceptive calendar, this data pertains to the 5 years prior to the survey. Using these 13 data is not considered feasible since a) there is only a small number of first births in that 14 interval (fewer than 140) and b) the recency of the births would severely constrain our 15 ability to make inference particularly for older marriage cohorts. About 57% of sexually 16 active women in the MDHS have reported not switching their contraceptive method within 17 the past 5 years. This is an important observation which validates the assumption that 18 women in Moldova are unlikely to switch their contraceptive method. 19 20 This analysis defines low contraceptive confidence for women who reported using a 21 traditional method (22% of women use either withdrawal or periodic abstinence), moderate 22 contraceptive confidence for those using a modern reversible method (e.g. pill, condom, 23 IUD, constituting 36%) and high contraceptive confidence for women using a permanent 24 reported not using any method: contraceptive confidence for these women cannot be 2 observed. We retain these women in the analysis however, since their abortion history is still 3 important in a context where abortion is normative fertility control behaviour. We include 4 two controls relevant to contraceptive behaviour: the month and year of first method use 5 and another variable measuring the previous method discontinued. 6 To capture the latent effect of abortion propensity, the analysis uses abortion history 7 as a proxy measure. Unfortunately, the MDHS has not collected any data on abortion 8 attitudes. We therefore use the proportion of pregnancies a woman has terminated. A 9 simple count is inadequate since older women have greater exposure to multiple abortions, 10 which may introduce bias. Using the proportion of pregnancies aborted overcomes this 11 problem. Other than recall problems inherent in cross-sectional surveys, any deliberate 12 under-reporting of abortion in post-Socialist countries is very low 22, 28 . Contraception and 13 abortion are often seen complementary in the Moldovan context-women report that the 14 use of ineffective methods (such as withdrawal) combined with frequent recourse to 15 abortion is a normative fertility control technique especially for traditional method users. An 16 interaction between contraceptive method and abortion propensity is used to test the 17 differential effect of abortion on different levels of contraceptive confidence. 18 Another key predictor variable is marriage cohort intended to capture the changes in 19 first birth rate which is often determined by economic circumstances especially the 20 availability of housing 23, 27, 29 . The age range of women in the dataset (15-49) means that 21 there should be some caution when interpreting results for the oldest marriage cohort since 22 there will be some left censoring: this marriage cohort is specified covering a wider range 23 than others to ensure sufficient sample size. The model controls for other effects which 24 could potentially influence the decision to have a first birth, the ability of women to conceive 25 and socio-demographic characteristics. These include: age at marriage, level of education of 1 women, geographical region and place of residence. As with the key explanatory variables 2 some of these are proxy variables limited to information available at survey. For example, 3 the duration of the first marriage is used to estimate whether the woman was in a 4 continuous marital union prior to first birth and whether union dissolution or separation 5 occurred before the first birth. Other control variables were considered in the model as they 6 were thought to be relevant a priori (ethnicity, wealth index, religious affiliation, 7 employment type, seasonality of employment, receipt of family planning media), but were 8 found not to significantly improve the model fit. Statistical significance was assessed by the 9 use of the Likelihood Ratio (LR) test with significance at the 5% level. The model was 10 estimated in SPSS 19.0. 11 12

Results 13
The regression results adjusting for relevant confounders and control variables are 14 presented for three selected effects (i) marriage cohorts, (ii) contraceptive confidence and 15 (iii) abortion propensity. The final model is presented in supplementary Table a1. Due the 16 interaction terms and time dependency specified in the model, it is difficult to interpret 17 coefficient directly, in particular the assessment of statistical significance of overall 18 probabilities. We therefore use this model to generate estimated survival curves and 19 cumulative hazards, and report the cumulative hazard of first birth at 12, 24 and 36 months 20 after marriage as a summary statistic in Table 1 as well as cumulative survival curves for each 21 main variable examined. In the tables, to allow the reader to assess significant effects, we 22 present confidence intervals adjusted for pairwise comparisons at the 5% level: the non-23 overlap of these intervals can be interpreted as a difference which is significant at the 5% 24 level. 25

Marriage cohorts 2
The adjusted hazard rate of a first birth for each duration since marriage is estimated for 3 different marriage cohorts. The results are shown in the form of survival plots (Figure 1), 4 truncated at 36 months for visual clarity. The survival plot indicates the proportion of 5 women yet to have first birth at month t following marriage. We also report the cumulative 6 hazard of first birth at 12, 24 and 36 months after marriage as a summary statistic in Table 1  7 a). overlaps with the most recent cohort after 24 months which suggests the propensity for 20 early transition to motherhood among recently married women. That said, the overall 21 probability of having a birth remains relatively constant-for instance 3 years following 22 marriage the later cohorts have attained the same proportion having had a birth as the pre-23 Socialist marriage cohorts. This is largely due to recuperation effect 2 to 3 years following 24 This is also reflected in the cumulative hazard, with the hazard among the pre-3 independence cohorts at 41%, 75% and 86% for 12, 24 and 36 months respectively. 4 However, there is a considerable fall in the cumulative hazard for the 1995-99 and 2000 5 marriage cohorts, indicating increasing delay of first birth following the collapse of Socialism, 6 but overall Moldovan women have a consistently high probability of becoming mothers. 7 8

Contraceptive confidence 9
The estimated survival curve for each level of contraceptive confidence is presented in 10 method is recorded at survey), the survival curve for high contraceptive confidence is the 20 highest, indicating the slowest transition to first birth in this group. Compared to women 21 with a low contraceptive confidence, the first birth rate is higher for women with a 22 moderate contraceptive confidence. The survival curve for high contraceptive confidence is 23 comparable to those of the low confidence group until 24 months following marriage 24 (indeed there is no detectable statistically significant difference at this point), when there is 1 a rapid fall in the proportion of women yet to have first births. This indicates that, in general, 2 low contraceptive confidence is associated with a low hazard of a first birth and hence 3 longer duration between marriage and first birth. On the other hand, an increase in 4 contraceptive confidence is associated with an increased hazard of a first birth, which clearly 5 suggests rapid transition to motherhood among women with high confidence. 6 7

Abortion 8
The estimated survival curve of first birth for women with low contraceptive confidence is 9 presented in Figure 3, which examines the association between low contraceptive 10 confidence and abortion propensity. In general, the proportion of women yet to have a first 11 birth is high for women with no abortion propensity, and the survival curves are lower for 12 women with low and moderate abortion propensity. Table 1 c) presents the estimated 13 cumulative hazard of first birth. Broadly, we see that the probability of having a first birth is 14 low for women with no abortion propensity. However, the cumulative hazard of first birth is 15 significantly higher at 12, 24 and 36 months among low and 12 and 24 months moderate 16 abortion users following marriage. This suggests that overall women who were prepared to 17 use abortion-at least partially have a shorter interval between marriage and first birth. The 18 survival curve for women with a high abortion propensity is roughly comparable or slightly 19 lower than women with no abortion propensity. We cannot detect an effect for high 20 abortion prevalence. Indeed there is some evidence of an attenuation in the higher 21 cumulative hazard of first birth at higher abortion levels: at 12, 24 and 36 months the 22 cumulative hazard is lower for moderate and high abortion users than women with a low 23 abortion propensity. 24

Key points
• Contraceptive confidence influences the duration between marriage and first birth in Moldova • There is a distinct effect of abortion on contraceptive confidence: the availability of abortion tends to increase women's contraceptive confidence • The effect of macro-economic shocks and social transitions are evident on marriage cohort specific first birth rates

Strengths and limitations of this study
• Study uses a nationally representative survey • Use of regression analysis disentangles net effects of related contraceptive and abortion behaviour • Use of retrospective data means necessitates reliance on proxy measures. Over the last two decades, many countries in Eastern Europe have experienced an 3 unprecedented decline in fertility with a Total Ffertility Rrate rates either at or below TFR 1.3 4 children per woman 1, 2 . Economic uncertainty and high male out-migration partly explain the 5 stagnant low fertility trends, although recent data show gradual recovery of fertility rates in 6 some countries 1,3 . Many women in Moldova tend to control their fertility by using 7 traditional contraceptive methods or induced abortions since modern method access is 8 limited 4, 5, 6 . This research focusses on Moldova where abortions are widely practised and 9 often accepted as a birth control method. 10 The dynamics of contraceptive use including discontinuation rates, switching and 11 method efficacy is widely acknowledged in demographic research 7, 8, 9, 10, c . However, the 12 confidence which women have in their contraceptive method and the effect it has on 13 fertility behaviour is under-researched. Contraceptive confidence is an hypothesis which 14 explains timing of childbearing resulting from the perceived efficacy of contraceptive 15 methods, but there is little modern literature 11 and much work examines older demographic 16 data 12, 13 . Theoretically, women who use less effective contraceptive methods (traditional 17 methods) have low contraceptive confidence, since their method is likely to fail. These 18 women tend to space their fertility as a means to limit their intended family size 12 . In 19 contrast, women who use effective (modern) contraceptives have a high degree of 20 confidence that these methods will not fail. This has prompted women to compress their 21 fertility into shorter periods 14,15 . 22 While previous studies have addressed second and later birth intervals, the 23 demographic landscape of Europe has undergone unprecedented changes in recent decades 24 driven mostly by changes in the relationship between partnership formation-particularly 25 marriage-and childbearing 1,16 . These trends are gradually emerging in Moldova signalling 26  have been a number of other explanations for changing fertility across Eastern Europe (for 7 example, more orthodox economic factors), and the cause is still debated among 8 demographers and dependent on context 19,20 . Therefore, any analysis exploring fertility 9 behaviour should account for marriage cohort as an important control variable, albeit not 10 one that can offer a complete explanation of observed trends. 11 We note that the pattern of union formation is an exceptionally complex 12 demographic process 11 . As well as the control variables we are able to include, there will 13 typically be significant variation in behaviour that are important but not captured by the 14 type of representative sample survey we employ. Therefore, while we are able to describe 15 part of the effects on first birth, this analysis should not be interpreted as a complete 16 picture. 17 In the Moldova, traditional methods are still widely used: about 26% of the 18 contraceptive methods used in Moldova are traditional 20,21 . This is considerably higher than 19 observed even in other former-Soviet Socialist countries (Latvia 8.7%, Hungary 9.0% and 20 Bulgaria 15.7%) 4 . Moldova therefore lends itself to examining the differential effects of  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59 1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59   From the original MDHS sample of 7,440 women, 1,884 women were excluded since 11 they were never married and 74% of these reported having never had sex. In addition, 179 12 women who had premarital births (2.4%) were excluded since the terminal event (first birth) 13

Marriage cohorts 2
The adjusted hazard rate of a first birth for each duration since marriage is estimated for 3 different marriage cohorts. The results are shown in the form of survival plots (Figure 1), 4 truncated at 36 months for visual clarity. The survival plot indicates the proportion of 5 women yet to have first birth at month t following marriage. We also report the cumulative 6 hazard of first birth at 12, 24 and 36 months after marriage as a summary statistic in Table 1  7 a). overlaps with the most recent cohort after 24 months which suggests the propensity for 20 early transition to motherhood among recently married women. That said, the overall 21 probability of having a birth remains relatively constant-for instance 3 years following 22 marriage the later cohorts have attained the same proportion having had a birth as the pre-23 Socialist marriage cohorts. This is largely due to recuperation effect 2 to 3 years following 24  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  This is also reflected in the cumulative hazard, with the hazard among the pre-3 independence cohorts at 41%, 75% and 86% for 12, 24 and 36 months respectively. 4 However, there is a considerable fall in the cumulative hazard for the 1995-99 and 2000 5 marriage cohorts, indicating increasing delay of first birth following the collapse of Socialism, 6 but overall Moldovan women have a consistently high probability of becoming mothers. 7 8

Contraceptive confidence 9
The estimated survival curve for each level of contraceptive confidence is presented in 10 method is recorded at survey), the survival curve for high contraceptive confidence is the 20 highest, indicating the slowest transition to first birth in this group. Compared to women 21 with a low contraceptive confidence, the first birth rate is higher for women with a 22 moderate contraceptive confidence. The survival curve for high contraceptive confidence is 23 comparable to those of the low confidence group until 24 months following marriage 24  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  (indeed there is no detectable statistically significant difference at this point), when there is 1 a rapid fall in the proportion of women yet to have first births. This indicates that, in general, 2 low contraceptive confidence is associated with a low hazard of a first birth and hence 3 longer duration between marriage and first birth. On the other hand, an increase in 4 contraceptive confidence is associated with an increased hazard of a first birth, which clearly 5 suggests rapid transition to motherhood among women with high confidence. 6 7

Abortion 8
The estimated survival curve of first birth for women with low contraceptive confidence is 9 presented in Figure 3, which examines the association between low contraceptive 10 confidence and abortion propensity. In general, the proportion of women yet to have a first 11 birth is high for women with no abortion propensity, and the survival curves are lower for 12 women with low and moderate abortion propensity. Table 1 c) presents the estimated 13 cumulative hazard of first birth. Broadly, we see that the probability of having a first birth is 14 low for women with no abortion propensity. However, , the cumulative hazard of first birth 15 is significantly higher at 12, 24 and 36 months among low and 12 and 24 months moderate 16 abortion users following marriage. This suggests that overall women who were prepared to 17 use abortion-at least partially have a shorter interval between marriage and first birth. The 18 survival curve for women with a high abortion propensity is roughly comparable or slightly 19 lower than women with no abortion propensity. We cannot detect an effect for high 20 abortion prevalence. Indeed there is some evidence of an attenuation in the higher 21 cumulative hazard of first birth at higher abortion levels: at 12, 24 and 326 months the 22 cumulative hazard is lower for moderate and high abortion users than women with a low 23 abortion propensity.  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59