Fertility intentions and outcomes

Advances in Life Course Research 23 (2015) 14–28 Contents lists available at ScienceDirect Advances in Life Course Research journal homepage: www.elsevier.com/locate/alcr Fertility intentions and outcomes Implementing the Theory of Planned Behavior with graphical models Letizia Mencarini a,b,*, Daniele Vignoli c, Anna Gottard c a b c University of Turin, Italy Collegio Carlo Alberto, Italy University of Florence, Italy A R T I C L E I N F O A B S T R A C T Article history: Received 9 February 2014 Received in revised form 2 October 2014 Accepted 15 December 2014 This paper studies fertility intentions and their outcomes, analyzing the complete path leading to fertility behavior according to the social psychological model of Theory Planned Behavior (TPB). We move beyond existing research using graphical models to have a precise understanding, and a formal description, of the developmental fertility decisionmaking process. Our findings yield new results for the Italian case which are empirically robust and theoretically coherent, adding important insights to the effectiveness of the TPB for fertility research. In line with TPB, all intentions’ primary antecedents are found to be determinants of the level of fertility intentions, but do not affect fertility outcomes, being pre-filtered by fertility intentions. Nevertheless, in contrast with TPB, background factors are not fully mediated by intentions’ primary antecedents, influencing directly fertility intentions and even fertility behaviors. ß 2014 Elsevier Ltd. All rights reserved. Keywords: Fertility intentions Theory of Planned Behavior Graphical models Italy 1. Introduction The study of fertility intentions has become central in the discussion of fertility rates in developed countries, under the realistic assumption that, in an almost perfect contraceptive regime, having a child is a result of a reasoned, although imperfect, decision. Whereas fertility intentions have been a central theme in demographic research for some time, it has received renewed attention in recent years since intentions are now frequently analyzed in the framework of Theory Planned Behavior (hereafter, TPB), a general psychological theory concerning the link between attitudes and behavior (Ajzen, 1991, 2005; Ajzen & Fishbein, 1980; Fishbein & Ajzen, 2010). Billari, Philipov, and Testa (2009), as well as Ajzen and * Corresponding author at: University of Turin, Italy. E-mail address: letizia.mencarini@unito.it (L. Mencarini). http://dx.doi.org/10.1016/j.alcr.2014.12.004 1040-2608/ß 2014 Elsevier Ltd. All rights reserved. Klobas (2013), specifically discuss the possible application of TPB in the fertility domain. Fertility outcomes, according to TPB, are seen as depending directly on fertility intentions, which in turn depend directly on attitudes (related to the perceived benefits and/or costs of reproduction), subjective norms (related to the social approval of behavior from relevant others), and perceived behavioral control. Possible constraints can further intervene from the time the fertility intention was formed and the subsequent behavior (such as a disruption of the couple’s relationship or changes in individuals’ health conditions or job status). This multi-factor paradigm is expected to depend on several background factors as well (such as socioeconomic and demographic factors). Whether TPB is a valid framework for analyzing human fertility is, however, a hotly debated issue (Ajzen, 2011; Barber, 2011; Klobas, 2011; Liefbroer, 2011; Miller, 2011; Morgan & Bachrach, 2011; Philipov, 2011) and, apart from the theoretical debate, there are only few attempts to test the TPB L. Mencarini et al. / Advances in Life Course Research 23 (2015) 14–28 framework in its full complexity (e.g., Dommermuth, Klobas, & Lappegård, 2013; Kuhnt & Trappe, 2013). In this study, we aim to analyze the complete path leading to fertility behavior, within the explanatory framework of TPB and considering the most common background variables (i.e., determinants) of fertility. We move beyond existing research developing a unified empirical strategy to test the validity of TPB. Through graphical models, we provide a precise understanding, and a formal description, of the developmental fertility decision-making process by studying the dependencies among all the factors involved in the TPB, on the basis of their joint distribution. In our application, variables are partitioned into a sequence of blocks: we can distinguish between pure response variables (in the last block), pure background variables (in the first block), and intermediate variables, which are responses for variables in previous blocks and explanatory for the subsequent variables. The partial ordering among the variables into blocks is fully derived by the TPB. The method, while not materially different from more conventional ones, does provide a useful conceptual fit to the TPB and facilitate its evaluation. Our study focuses on Italy, a country for which the use of TPB in fertility research is a novelty. We rely on a specific module questionnaire, within the 2003 Italian GGS survey, that was designed to collect the relevant dimensions of the TPB (Vikat et al., 2007). Then, using the 2007 follow-up of that survey, we look at the subsequent fertility behavior, thus completing the study of the whole process leading up to the decision to have a child. 2. Theoretical background on the study of fertility intentions and realizations 2.1. The Theory of Planned Behavior as a theoretical framework for the fertility decision-making process According to the TPB, which is an extension of the earlier Theory of Reasoned Action (Fishbein & Ajzen, 1975, 2010), intentions are the immediate antecedents of corresponding behavior. This hypothesis is supported by several systematic reviews of the empirical literature, and strong intention–behavior correlations are also observed in the fertility domain (Ajzen, 2010; Billari et al., 2009). As Ajzen and Klobas (2013) advocate, however, a concern in applying the TPB in the study of human fertility is defining an appropriate behavioral criterion. In fertility research, having a child, is commonly described as a behavior. However, strictly speaking, a child birth is an outcome of specific antecedent behaviors (e.g., having sex, not using a contraceptive, using artificial reproductive technology, etc.) that (may) result in a pregnancy. From this perspective, having a child is an outcome or behavioral goal, rather than a behavior, which might result in attainment of the goal. Central to this discussion is the implicit or explicit assumption that, at least in developed countries with readily available contraception, having a child is the result of a reasoned decision (e.g., Goldstein, Lutz, & Testa, 2004; Mills, Mencarini, Tanturri, & Begall, 2008; Testa & Grilli, 2006). In almost perfect contraceptive regimes, in fact, the difference between fertility as a behavior and fertility as a 15 behavioral goal is as narrow as it can be. Of course, individuals have greater control over performance of a behavior than they have over attaining a goal the behavior is intended to produce. Ajzen and Klobas (2013) describe how the TPB can be used to model fertility decision-making process: when people formulate their intentions to have a(nother) child, they rely on three conceptually distinct, but interrelated, primary antecedents of fertility intentions: attitudes, subjective norms, and perceived behavioral control. These antecedents represent the most important predictors of fertility intentions. The ‘‘attitudes toward the behavior’’, which can be favorable or unfavorable, are ‘‘readily accessible or salient beliefs about the likely consequences of a contemplated course of action’’ (Ajzen, 2010). In the case of fertility decision-making, individuals would be expected to reflect on their attitudes about having a child before forming their fertility intentions. Such attitudes are a person’s internal evaluation that having a child will have positive or negative (i.e., desirable or non-desirable) consequences for her/him. The ‘‘subjective norms’’ are related to the perceived normative beliefs and expectations of relevant referent groups or individuals who exert social pressure to perform or avoid the behavior. In the case of fertility intentions, individuals would be expected to consider subjective norms for having a child; i.e., the individual’s perception of the psychological support of or normative pressure on her/his fertility behavior from members of her/his close social circle. Finally, individuals are assumed to take into account factors that may promote or hinder their ability to perform the behavior, and these salient control beliefs lead to the formation of ‘‘perceived behavioral control’’. This refers to the perceived capability of performing the behavior. Because many behaviors pose difficulties in execution, it is useful to consider perceived behavioral control over having a child in addition to intentions. Like attitudes and subjective norms, perceptions of behavioral control follow consistently from readily accessible beliefs about resources and obstacles that can facilitate or interfere with the ability to have a child, such as income or wealth, labor force status, and education (Billari et al., 2009). The power of each control factor to facilitate, or inhibit, behavioral performance is expected to contribute to perceived behavioral control in direct proportion to the subjective probability that the control factor is present in each person (Ajzen, 2010). According to Ajzen and Klobas (2013) fertility intentions are also expected to result in having, or not having, a child to the extent that people are in fact capable of attaining their goals, i.e., to the extent that they have actual control over having a child. Actual behavioral control is thus hypothesized to moderate the effect of intention on behavior. Because actual behavioral control is identifiable with difficulty, perceived control is often used as a proxy for actual control, under the assumption that perceptions of control reflect actual control reasonably well. Demographic research directed toward explaining or predicting fertility intention within the reasoned action tradition of the TPB has focused primarily on the intention to have a child relative to the intention to not have a child 16 L. Mencarini et al. / Advances in Life Course Research 23 (2015) 14–28 Fig. 1. A schematic presentation of the Theory of Planned Behavior for fertility decision-making. Adapted from Ajzen and Klobas (2013). (Billari et al., 2009; Jaccard & Davidson, 1975; Jorgensen & Adams, 1988) and on the timing of the intentions (intentions in the short run versus intentions in long run, Dommermuth, Klobas, & Lappegård, 2011), often disregarding – mainly because of lack of suitable data – the complete path leading to fertility behavior. A schematic representation of the TPB applied to fertility research is shown in Fig. 1. TPB does not discount the importance of background factors that can influence behavior indirectly. They are selected by a content-specific theory, which can complement the TPB ‘‘by identifying relevant background factors and thereby extending understanding of a behavior’s determinants’’ (Ajzen, 2010). Therefore, a number of well-established factors studied in demographic research, such as age, parity, or education, are treated as external variables. Under ideal conditions, in the operationalization of the TPB the background factors should affect only the primary antecedents of fertility intentions, and should not have a direct impact on the intentions themselves, or on fertility outcomes. It is worth noting that it is difficult to empirically disentangle actual behavioral control and pure background factors. What, with detailed survey data, can be identified is the effect of actual enablers and constraints which intervene between the moment of declaration of fertility intentions and the time of the child birth. 2.2. Empirical evidence on the determinants of fertility intentions and realizations According to the TPB, the distinction between attitudes, norms, and perceived behavioral control should completely filter the role played by background factors on fertility intentions, which will in turn determine the subsequent realization. It is therefore crucial to choose a set of background factors which have been proven to influence both fertility intentions and their realization. There is a plethora of empirical research focusing on the determinants of fertility intentions suggesting that they depend on several demographic, socioeconomic, and gender-related factors (e.g., Berrington & Pattaro, 2013; Cavalli & Klobas, 2013; Kapitány & Spéder, 2012; Mills et al., 2008; Neyer, Lappegard, & Vignoli, 2013; Spéder & Kapitány, 2014; Thomson, 1997; Thomson, McDonald, & Bumpass, 1990; Vignoli, Rinesi, & Mussino, 2013). In contrast, the literature investigating the correlates of the realization of fertility intentions is scarce, mainly due to a severe lack of appropriate longitudinal data. It has been suggested that the factors affecting fertility intentions are both demographic and socioeconomic. Of the purely demographic factors, parity and the woman’s age play crucial roles in the definition of fertility intentions (Berrington, 2004; Bongaarts, 1992, 2001; Liefbroer, 2009; Morgan, 1982; Noack & Østby, 1985; Rinesi, Pinnelli, Prati, Castagnaro, & Iaccarino, 2011; Thomson, 1997). Generally, the documented findings show that there is an inverse relationship between fertility intentions and parity (Bühler, 2008; Thomson, 1997). Positive fertility intentions – i.e., the intentions to have a child – also seem to be less frequent among older women. The effect of the type of union has also been widely investigated. Liefbroer (2009) showed that married women have higher average fertility intentions than those who do not have a partner or who are cohabiting. Accordingly, Régnier-Loilier and Vignoli (2011) have found that, in Italy, cohabiting couples want fewer children than married couples (however, in France, such effect has not been found; Régnier-Loilier & Vignoli, 2011; Toulemon & Testa, 2005). The role of education was emphasized in a crosscountry study by Heiland, Prskawetz, and Warren (2008). In many European societies, higher educational levels were associated with the intention for a greater number of children. The positive association between educational level and fertility intentions was also confirmed by a study on France (Toulemon & Testa, 2005) and by a study on Bulgaria and Hungary (Philipov, Spéder, & Billari, 2006). Compared to women with less education, highly educated women displayed better labor market opportunities and earnings, as well as greater bargaining power within the couple, which encourages a more equal L. Mencarini et al. / Advances in Life Course Research 23 (2015) 14–28 division of housework and childcare between partners, and which could in turn facilitate fertility intentions (Mills et al., 2008). Among the relatively few studies which have looked at the transformation of fertility intentions into realization, demographic factors appear to play a pivotal role. In particular, woman’s age and parity are crucial (e.g., Noack & Østby, 2002; Quesnel-Vallée & Morgan, 2003; Rinesi et al., 2011): postponing motherhood results in having fewer children than originally planned. Moreover, the greater the distance between the actual and expected number of children, the faster the transition toward childbearing in a short period (Symeonidou, 2000; Thomson et al., 1990). The type of union is also important. Married couples are more likely to realize their intention to have a/nother child in the United States (Quesnel-Vallée & Morgan, 2003; Schoen, Astone, Kim, Nathanson, & Fields, 1999), Italy, and France (Régnier-Loilier & Vignoli, 2011). The fertility intentionrealization gap is smallest for highly educated women (Rinesi et al., 2011; Toulemon & Testa, 2005). Finally, the effect of gender roles seems to vary in different contexts: in Greece, less traditional women have greater difficulties than more traditional women in realizing their positive fertility intentions (Symeonidou, 2000); while in other contexts, such as Sweden, the trend is reversed (Thomson, 1997). 3. Data 3.1. Italian Gender and Generation Survey We use the Italian Gender and Generation Survey and its corresponding follow-up survey. The Italian variant of the GGS is a prospective and retrospective survey conducted by the Italian National Statistical Office (Istat) in 2003, which is called Family and Social Subjects, or GGSFSS (2003). The follow-up survey, which looked at critical points in life histories from a gender perspective, was jointly conducted by Istat and the Ministry of Labour in 2007. GGS-FSS (2003) has a sample of about 24,000 households and 50,000 individuals of all ages (with a nonresponse rate of 17.7%). Out of them, almost 32,000 are aged 18–64. The follow-up includes about 10,000 interviews with people aged 18–64. In this wave, the overall non-response rate was 48.6%. Individuals’ intentions to have a child within the next three years were surveyed using the following question: ‘‘Do you intend to have a child in the next three years?’’. The four possible answers were: ‘‘definitely not’’, ‘‘probably not’’, ‘‘probably yes’’, and ‘‘definitely yes’’. Limiting the question about childbearing intention to a foreseeable time frame avoided some of the problems associated with the surveying of intentions. Questions about intentions, that cover a foreseeable time period, are ‘‘in close temporal proximity to the prospective behavior’’ (Ajzen & Fishbein, 1973) and are generally considered to be better predictors of behavior (Billari et al., 2009; Philipov, 2009). They allow researchers to make inferences based on a person’s current status about what economic, institutional, and familial conditions are crucial in her/his decision process to have a/nother child. 17 Our selected sub-sample consisted of 2871 women, aged 18–49, married or cohabiting with a partner.1 We focus on them because the battery of questions related to TPB was asked only to individuals in co-residing couples. Overall, linking the two Italian GGS waves allowed us to assess whether the fertility intentions were subsequently realized. We covered the period of 2003–2007, during which the births2 of 368 children were registered. The association between the birth of a child as the outcome of interest and prior intentions turned out to be particularly strong at the extremes: the stronger the intention to have or not have children, the greater or the lower the observed proportion of respondents who realized this intention.3 We found that negative fertility intentions are a potent predictor of subsequent fertility behavior. By contrast, positive fertility intentions tended to overestimate fertility realizations: about 40% of the Italian respondents who firmly stated the intention to have a child in the following three years did not achieve their goal. Incidentally, these results are relatively consistent with those of a handful of other studies for Italy based on different data sources (e.g., Rinesi et al., 2011). 3.2. The relevant variables and dimensions for the TPB model As in all of the Gender and Generation Surveys, the 2003 Italian FSS survey included a battery of questions needed to implement the TPB (Vikat et al., 2007). Ten items were utilized to characterize attitudes toward having a child. Each of these items was introduced by the question: ‘‘If you were to have a/nother child within the next three years, would it be better or worse in relation to. . .’’. Among the possible responses were: ‘‘much better’’, ‘‘better’’, ‘‘neither better nor worse’’, ‘‘worse’’, and ‘‘much worse’’. Subjective norms were measured through three questions.4 The respondents were asked to rate the extent to which they agree that different groups of people think they should have a/nother child. All three items were introduced by the following question: ‘‘If you were to have a child in the next three years, to what extent would the following persons agree with your choice?’’. Among the 1 In this study, we have deliberately chosen to consider the path leading to fertility behavior only for women. Reproductive behavior in advanced societies is most of time a joint decision of the two partners, therefore conducting a study at the couple level would have been of substantial interest. In our research framework, we could have estimated all the graph models specifically for men, or considered men and women samples pooled together. However, in the first case this would have produced the double of figures and graphs; in the second case to assess properly gender issues and looking at couples, we should have tested several interactions complicating the graphical representation of the results, and, moreover, this would also have implied considering a variable that is not embedded within the TPB – i.e., the couple’s agreement over fertility intentions. 2 These births are live births. Because of lack of specific information we cannot consider subfecundity (i.e., miscarriages, still births and induced abortions). 3 See Régnier-Loilier and Vignoli (2011) for a detailed discussion on the intentions’ predictive strength using the same data. 4 The Italian FSS-GGS omitted one important item included among the subjective norms according to the GGS guideline (Vikat et al., 2007), namely ‘‘other relatives’’. 18 L. Mencarini et al. / Advances in Life Course Research 23 (2015) 14–28 possible responses were: ‘‘would strongly agree’’, ‘‘would agree’’, ‘‘would neither agree nor disagree’’, ‘‘would disagree’’, ‘‘would strongly disagree’’. Finally, the survey included seven items intended to capture perceived behavioral control. The following question was posed for each of the item: ‘‘the decision about whether to have children can depend on various situations. How much would your decision about whether to have a child in the next three years depend on. . .’’. Among the possible answers were: ‘‘a lot’’, ‘‘quite lot’’, ‘‘a little’’, and ‘‘not at all’’. In the case of perceived behavioral control, we reversed the scale, because this made it easier to show the possible positive effect of the perceived ability to overcome constraints with a positive coefficient in the regression model. We used factor analysis to verify whether the items available in the used survey acted as valid and reliable measures of the TPB variables (Billari et al., 2009; Dommermuth et al., 2011).5 We tested both a three-factor solution (the proposed factors were attitudes, subjective norms, and perceived behavioral control) and a four-factor solution (which allowed for attitudes to fall into two groups), as was done in Billari et al. (2009) and Dommermuth et al. (2011). Four factors can be identified (see Table A1 in the Appendix, where also the list of the used items is reported). Two can be reasonably interpreted as attitudes factors, one as a measure of subjective norms, and one as a measure of perceived behavioral control. The first of these factors is hence called ‘‘negative attitudes’’, as it represents beliefs about the costs or personal losses associated with having a child, while the second is called ‘‘positive attitudes’’ because it represents beliefs about the benefits of having a child. 4. A chain graph model for implementing the TPB The TPB itself illustrates a temporal sequence for the process leading to the decision to have a child. The existence of such a sequence suggests the possibility of using the class of graphical models to gain a precise understanding of the developmental fertility decisionmaking process, as they allow us to study the conditional independence structure among all of the variables involved in this process, and depict this structure by a graph. Graphical models,6 developed as an extension to path analysis (Wright, 1921), are a class of multivariate models that are useful for studying, estimating, describing and visualizing the relationships among an entire set of variables of interest. Only few applications of this class of models have been recorded in social sciences (e.g., Berrington, Hu, Smith, & Sturgis, 2008; Gottard, 2007; Mohamed, Diamond, & Smith, 1998). A multivariate model is graphical whenever its conditional independence structure can be univocally depicted by a graph. A graph G = (V, E) consists of two finite sets: a set V for nodes and a set E collecting edges between nodes. The edges in a graph 5 We used alpha factor analysis of the correlation matrix and then performed a rotation of the loading matrix through oblimin criteria. 6 See Lauritzen (1996) and Edwards (2000) for a comprehensive introduction of graphs and graphical models. can be undirected (lines), or directed (arrows). In graphical models, nodes represent variables, and the absence of a connection between two nodes represents a conditional independence. Graphs are therefore utilized to give a theoretically rigorous but intuitively easy to understand representation of the complex relationships among variables, on the basis of their joint distribution. These relationships are described in terms of conditional independence, which is the key concept of graphical models. Two variables, X and Y, are conditionally independent given a third variable Z, denoted by X E YjZ, when controlling for Z, X does not provide additional information on the distribution of Y, and vice versa. A conditional independence statement is visualized in the graph by the absence of a connection between two nodes. In this study we use chain graph models, which are the most appropriate to empirically implement the temporal sequence of the fertility decision-making process. They are in fact graphical models associated with a chain graph (Lauritzen & Wermuth, 1989): i.e., a graph with both directed and undirected edges. These graphs, also called block-recursive graphs, are particularly useful when, as in our case of the fertility decision-making process, variables admit a partial ordering on the basis of subject matter considerations, as hypothesized by the TPB. Variables are then partitioned into blocks. Those variables belonging to a same block are considered to be of equal standing, while those belonging to different blocks can be joined by arrows, representing a directional association. Consequently, it is possible to distinguish between pure response variables (in the last block), pure explanatory/background variables (in the first blocks), and intermediate variables, which are responses for variables in previous blocks and explanatory for the subsequent variables. Here we adopt the Lauritzen–Wermuth–Frydenberg (LWF) class of Markov properties, which establish how conditional independencies can be deduced by a graph (see Lauritzen and Wermuth (1989), Frydenberg (1990), and Borgoni, Berrington, and Smith (2012), for a detailed presentation on how read conditional independence in a LWF chain graph). To better understand how conditional independence can be read from a chain graph, it is helpful to look at Fig. 2, which shows a chain graph consisting of five blocks, as in our case. The first block (a) on the left collects the two nodes corresponding to the two pure explanatory variables X1 and X2. The blocks (b), (c) and (d) contain intermediate variables: each of them is explanatory for variables in the block on the right, response for those on the left and concomitant for those in the same block. Finally, the last block (e) contains the variables X7 and X8, which are studied only as responses, given all the other variables. For example, in Fig. 2, the lack of the edge between nodes 1 and 2 corresponds to the marginal independence statement X1 E X2, as no previous blocks are present. On the other hand, the absence of the edge between nodes 3 and 4 indicates that X3 E X4jX1, X2. In a LWF chain graph model as adopted here, estimates for the parameters of the joint distribution of the variables was obtained via maximum likelihood. The multidimensional problem has been simplified by factorizing the joint distribution according to the block structure of the graph, L. Mencarini et al. / Advances in Life Course Research 23 (2015) 14–28 19 Fig. 2. Example of a chain graph with five blocks. into a sequence of univariate models, as suggested by Cox and Wermuth (1996). Each univariate model has been specified according to the nature of the response, where all of the variables in the same and in the previous blocks are considered as explanatory. The edges selection was achieved using a stepwise procedure which compares the model for the reduced graph with the complete graph by means of the Likelihood Ratio test (with the significance level set at 0.05). In the setting of the chain graph, the TPB guides us to put variables into blocks. The sequence of the fertility decisionmaking process is produced by ‘‘background variables’’ (block a), ‘‘perceived behavioral control,’’ ‘‘subjective norms,’’ ‘‘positive and negative attitudes’’ (block b), ‘‘fertility intentions’’ (block c), ‘‘actual constraints’’ (block d), and ‘‘fertility outcome’’ (block e); see Table A2. The background variables of the first block (a) were selected based on evidence from the literature, as outlined in Section 2.2, as well as based on the peculiarities of the Italian context (e.g., Dalla Zuanna, 2001; De Rose, Racioppi, & Zanatta, 2008; Mencarini & Tanturri, 2006; Rinesi et al., 2011). They are: number of children, woman’s age, duration of the couple’s relationship, type of couple, woman’s and man’s educational levels and employment situations, gender arrangements between the partners, religiosity, number of siblings,7 macro-region of residence, municipality size. The dependence structure among the background variables has not been studied because their association is far outside of the scope of the paper. The variables in the second block (b) are the variables at the core of the TPB. On the basis of the four latent factors previously identified (see Section 3.2), we constructed four derived variables.8 These derived variables are considered 7 The number of siblings is indeed the fertility of interviewer’s parents and is one of the key determinants of fertility intentions: it has been shown (see, for instance, Régnier-Loilier, 2006) that generally those who have some siblings have better memories about their childhood compared to lonely children, and therefore, on average, they tend to desire for themselves a higher number of offspring. 8 The derived variables were obtained by summing the original variables and were then considered as ordinal variables. We preferred derived variables to factor loadings as they are not latent variables, and can therefore be more easily inserted into the joint distribution of the chain graph model, while still giving similar information. See Cox (2008) for the properties of sum-derived variables. as intermediate variables, as they are dependent on the background variables and explanatory for the variables in the following blocks. As dependent variables, they were modeled by adopting a cumulative logit model for ordinal variables. For the variable in the third block (c), we estimated a cumulative logit model predicting fertility intentions.9 The fourth block (d) of ‘‘actual constraints’’ includes the binary variable for the disruption of the couple’s relationship, which is studied as a logit model. The inclusion of this block is in line with the TPB and highlights the importance of intervening factors between the time when fertility intentions are declared and the fertility outcome. These factors tend to be neglected by fertility studies that assume that the determinants of intentions also influence the subsequent behavior. Some problems or constraints, which inhibit the realization, can however arise after the moment when the intentions are expressed, such as unexpected conflicts between the partners (Régnier-Loilier & Vignoli, 2011). Finally, the last block (e) consists of the pure response variable – i.e., the fertility outcome – whose dependence on the variables in the previous blocks was estimated by adopting a logit model. It should be noted that an important logical element in the TPB is the proposed effect of actual behavioral control on perceived behavioral control. Fertility intentions are expected to be transformed into a behavior to the extent that people are capable of attaining their goals; i.e., to the extent they have actual control over having a child. Background factors and actual behavioral controls are theoretically not well discernible and, as result, there is a high likelihood that background factors are seen empirically to have a direct effect on intentions. We believe that the main difference between the two concepts is that people are able to assess their ‘‘perceived behavioral control’’ at the moment they form their intention while they are not able to foresee ‘‘actual constraints’’ after the intention, and therefore we have chosen the variables 9 Note that the intentions to a have a child are often considered in the literature as ‘‘parity-progression intentions’’ (Billari et al., 2009). However, our study had to use a very small-scale sample, and, as a consequence, we could not stratify our analysis by parity. 20 L. Mencarini et al. / Advances in Life Course Research 23 (2015) 14–28 within blocks accordingly (i.e., considering as actual constraints – block d – only those factors intervening between the two waves of the survey). 5. Structure of (in)dependence among the variables of TPB Overall, the analysis produced a large and interesting set of empirical findings, but our specific focus is on dependency structure among the variables, i.e., what is in/ dependent of/on what. We deliberately abstain to describe in details the direction of the associations we found among the variables involved in the TPB and we provide insights regarding the direction of the effects only when it appears to be relevant for the discourse (the detailed table of results is in Table A3 in the Appendix). However, it is worth nothing that the results from the graphical models are fully consistent with those of other studies on fertility intentions for the Italian case. Consistently with the theory, looking at Fig. 3 we can see that the primary antecedents of fertility intentions considered in the TPB (PAtt, NAtt, SubN, PBC) are correlated with one other. Only the ‘‘perceived behavioral control’’ (Pbc) is independent of the index standing for ‘‘subjective norms’’ (SubN), given the ‘‘positive’’ (PAtt) and ‘‘negative attitudes’’ (NAtt), as well as the background variables (Fig. 3). We can also discern which background variables influence the primary antecedents of fertility intentions, conditionally one to another. The ‘‘perceived behavioral control’’ depends on the number of children (Nch), the current gender division of housework in the family (CHD), and the woman’s age (AgeW). The ‘‘positive attitudes’’ dimension depends on the degree of religiosity (Rel), the number of children (NCh), and the duration of the couple’s relationship (CDur). The ‘‘negative attitudes’’ dimension depends on the number of children (Nch), the current division of housework (CHD), and the area of residence (Reg). The ‘‘subjective norms’’ dimension depends on the number of children (Nch), the number of siblings (Sib), the duration of the couple’s relationship (CDur), the woman’s age (AgeW), and the number of children (NCh). Overall, the only background factor influencing all the primary antecedents of fertility intentions is the number of children, while some of the background variables do not influence any of the primary antecedents of fertility intentions. This fact can be visualized in Fig. 3 by the fact that they stand alone: the man’s and the woman’s employment situations (ME, WE), the woman’s satisfaction regarding the housework division (SHD), the parents’ residential proximity (PRP), the municipality size (MunS), the type of couple (CTy). Fig. 3. Conditional independence graph of background variables (block a) and primary antecedents (block b). L. Mencarini et al. / Advances in Life Course Research 23 (2015) 14–28 21 Fig. 4. Conditional independence graph of background variables (block a), primary antecedents (block b), and fertility intentions (block c). Note: Arrows from block a (background) to block b (primary antecedents) are not shown for simplicity (they were displayed in Fig. 3). Corroborating the scheme of TPB, the level of fertility intentions (Fint) depends on all of the predicted primary antecedents of fertility intentions: (positive and negative) attitudes, subjective norms, and perceived behavioral control (Fig. 4). According to TPB, under ideal conditions and operationalization, the background factors should affect only the primary antecedents of fertility intentions, and should not have a direct impact on the intentions themselves. However, our empirical analysis does not validate this part of the theory, showing that some of the background variables also have a direct effect (this is also in line with Billari et al., 2009). In this respect, for some of the background variables, the TPB works: this is the case, for instance, regarding the couple’s education (CEd), as its influence on fertility intentions is mediated by subjective norms (SubN). By contrast, a direct influence of the background variable on fertility intentions has been found for several demographic factors, such as the number of previous children (NCh), the woman’s age (AgeW), and the duration of the couple’s relationship (CDur); for some socioeconomic factors, such as the woman’s employment status (WE) and the partners’ housework division (CHD); as well as for culturally driven factors, such as the degree of religiosity (Rel). Overall, these results confirm that stronger equality in terms of division of household labor favors positive fertility intentions, suggesting that a perception of fairness of gender arrangements facilitate Italian women’s plan to have a child in the next period, net the other factors involved (see also Mills et al., 2008). Moving to the constraints that may intervene between the time when people express their fertility intentions and the moment of the subsequent realization, we considered the likelihood of a disruption of the couple’s relationship (Fig. 5, blocks a–d). This event (CDis) is independent of the level of intentions (FInt) and of all primary antecedents of fertility intentions (PAtt, NAtt, SubN, PBC), given the background variables. Among the latters, the woman’s age (AgeW), the duration of the couple’s relationship (CDur), the type of couple (CTy), the woman’s employment status (WE), the woman’s level of satisfaction with the housework division (SHD), and the municipality size (MunS) have a significant impact on the couple’s disruption risk. Fig. 5 also offers evidence for the empirical validation of the TPB. As predicted, having a child (Child) is conditionally independent of attitudes (PAtt, NAtt), subjective norms (SubN), and perceived behavioral control (PBC); instead, these influence the previous step (the formation of intentions). In other words, the intentions (Fint) act as a filter between the primary antecedents of fertility plans and the subsequent behavior. As expected, both the likelihood of a disruption of the couple’s relationship (CDis) and the level of intentions (Fint) directly influence fertility outcomes (Child): the likelihood of having a child is higher when the fertility intentions are higher and when the couple remains together in the intervening period (see Table A3 in the Appendix). However, the fertility behavior is also directly influenced by some of the background variables, without being filtered by prior blocks. In particular, the most important determinants of having a child are the demographics, such as the woman’s age (AgeW), the number of children (NCh), and the duration of the couple’s relationship (CDur). This 22 L. Mencarini et al. / Advances in Life Course Research 23 (2015) 14–28 Fig. 5. Conditional independence graph of background variables (block a), primary antecedents (block b), fertility intentions (block c), actual constraints (block d), and fertility outcomes (block e). Note: Arrows from block a (background) to block b (primary antecedents) as well as from block a (background) to block c (constrains) are not shown for simplicity. has been already stated in previous research (e.g., Noack & Østby, 2002; Quesnel-Vallée & Morgan, 2003; Rinesi et al., 2011; Régnier-Loilier & Vignoli, 2011), although without an examination of the joint distribution of all of the variables involved in the TPB, as was done in this paper. Importantly, the couples’ gender role arrangements have a direct effect for making the step from fertility intentions to the subsequent behavior. We note that fertility realizations are naturally higher with stronger intentions, but they also depend on couple stability. The realizations are also influenced by demographic variables that are not prefiltered by preceding blocks (i.e., woman’s age, number of children, and duration of the relationship matter). This finding confirms that Italian reproductive behavior is strongly correlated with the demographics, likely being also drivers of subsequent infecundity and involuntary low fertility or childlessness. In sum, the application of graphical models gives interesting insight to verify if the TPB holds for Italy. The result is mixed, and can be summarized in Fig. 6. On one hand, the intermediate variables influence fertility outcomes only by intention (in the above scheme, there is not a direct link between intermediate variables and child). This finding supports the TBP. On the other hand, the direct links between background factors and fertility intentions, as well as between background factors and fertility outcomes, are not in line with the TPB. 6. Concluding discussion We have followed a common paradigm, expecting individuals to make their procreative choices intentionally. We relied on a framework built from the TPB (Ajzen, 1991) and a further adaptation of the TPB to the fertility decisionmaking process (Ajzen, 2010, 2011; Ajzen & Klobas, 2013). In this paper the TPB is operationalized including fertility behavior (i.e., the outcomes), while most previous research applied it for intentions only (Billari et al., 2009; Dommermuth et al., 2011). Moreover, we moved beyond existing research by representing all of the possible dependencies among the variables involved in the TPB, studied jointly, by means of a chain graph. Our findings yield new results for the Italian case which are empirically Fig. 6. Stylized summary of results. L. Mencarini et al. / Advances in Life Course Research 23 (2015) 14–28 robust and theoretically coherent, adding important insights to the effectiveness of the TPB for fertility research. Overall, this research represents a contribution to the study of contemporary fertility decision-making process from a life course perspective. In this respect, a first key finding is that attitudes, norms, and perceived behavioral control are distinct determinants of fertility intentions, even after adjusting for possible confounding background factors. And, even more relevant, a second finding is that none of these dimensions has a direct effect on fertility behavior: they are all pre-filtered by fertility intentions. Finally, as a third key finding, we illustrated that the background factors affect both fertility intentions and realizations and therefore are not fully mediated by fertility intentions primary antecedents. The first two findings are consistent with the theoretical predictions of the TPB and speaks for an effective and replicable application of TPB in fertility research. We posit that the distinction between attitudes, norms and behavioral control is a strategy that encourages a simplification of the complexity of factors leading to fertility behavior. Our study proposes an approach to the study of fertility decision making from a life course perspective as a key to understanding fertility behavior in modern societies. Importantly, given that similar survey instruments to the ones used in this paper have been implemented in the Generations and Gender Survey (GGS) for several other societies in Europe (Vikat et al., 2007), the approach is also replicable in a systematic manner for other contexts. Nevertheless, the third key finding poses a crucial problem for the operationalization of TPB because of the role of background factors, complicating the application of the TPB to fertility research. According to the TPB, under ideal conditions and operationalization, the background factors should affect only the primary antecedents of fertility intentions – namely, attitudes, subjective norms, and perceived behavioral control – and should not have a direct impact on the intentions themselves. Contradicting these assumptions, our findings indicated that, in the course of the fertility decision-making process, not all background factors are mediated by the fertility intentions’ primary antecedents. Net of the entire structure of (in)dependence that characterizes the variables involved in the TPB, some of the background factors directly influence fertility intentions, and others even influence fertility behaviors. The lack of independence among background factors, fertility intentions, and outcomes inhibits a complete validation of TPB for fertility research in the Italian setting. We located three practical potential reasons for this. First, a data issue can be at play: it is possible that the battery of questions provided by the Italian GGS survey do not fully capture the three conceptually separated dimensions proposed by TPB. When measurement of the antecedents of intentions is flawed, related background factors will compensate and have a direct effect on fertility intentions and, thus, on fertility outcome (Ajzen, 2010; Ajzen & Klobas, 2013; Dommermuth et al., 2013). Second, our research design can matter. For pragmatic reasons, due to the smallscale sample, we had to consider all parities jointly estimating the likelihood of having a child (net of the 23 parity level). Background factors could be parity-specific, however. We cannot exclude that a future study employing a sample large enough to stratify the analysis by parity will bring about different results. Third, a common problem of all the studies which have tried to empirically test the TPB (e.g., Dommermuth et al., 2011) regards the use of short panels, having only two waves. Because the primary antecedents and the fertility intentions were all collected at one point in time, the data failed to provide a rigorous test of TPB: there is no empirical basis for establishing that the primary antecedents are causally prior to intentions. A corollary is that some of the background variables used in the model specification may also be the product of fertility intentions (e.g., employment, gender role set). Future research should operationalize the TPB using longitudinal data with at least three waves in order to insert a time window not only between the expression of the fertility intentions and their subsequent outcome, but also between the surveying of the primary antecedents of intention and the surveying of the intentions themselves. A part from data-related reasons, a potential intriguing and insidious further explanation for the direct effect of background factors of fertility intentions is theoretical. Our findings may also venture the existence of further (psychological) factors which are unobserved, and therefore not built-in the three type of consideration – attitudes, subjective norms, and perceived behavioral control – predicted by the TPB. Nevertheless, the link between background factors and fertility outcomes does not seriously contrast the TPB, because ‘‘actual enablers and constraints’’ (placed between the moment of fertility intentions declaration and their subsequent realizations) is under-considered in the paper. Due to data limitations, we only considered the dissolution of the couple as a possible enabler. The arrays between background factors and fertility outcomes would probably be statistically irrelevant if we had considered additional intervening constraints before the realization of intentions. To conclude, it seems relevant to point out that the analysis could be alternatively carried out using the more common structural equation models (SEM) instead of graphical models. Both the classes of models are adequate for this analysis, even if they are not equivalent (for details, see Cox & Wermuth, 1993; Smith, Berrington, & Sturgis, 2009). Although less known, graphical models are a wellestablished class of multivariate models, with interesting probabilistic and computational properties, whose effectiveness and usefulness has still to be explored in social and demographic studies. The chain graphs utilized in this paper turns to be not distributionally equivalent to structural equation model. The advantage in using chain graphs, even if less known to a general public, is that they provide a clear interpretation of model parameters in terms of conditional independence. Such independence fully corresponds to the missing edges in the corresponding graph. Moreover, none latent variable is introduced in the model so that no additional distributional assumptions are required. As a matter of fact, conditional independence is the key to understand and interpret graphical models, which seems to us particularly suitable for assessing the TPB for the fertility decision-making process. 24 L. Mencarini et al. / Advances in Life Course Research 23 (2015) 14–28 Appendix See Tables A1–A3. Table A1 Factor analysis of items of perceived behavioral control, subjective norms, and attitudes regarding the intention to have a/another child within the next three years. Items Factor 1: Negative Attitudes Factor 2: Positive attitudes Factor 3: Subjective norms Factor 4: Perceived behavioral control If you were to have a/another child during the next three years, would it be better or worse in relation to. . . 0.58 - The possibility of doing what you want 0.55 - Your employment opportunities - Your partner’s job opportunities 0.30 - Your financial situation 0.59 0.42 - Your sexual life - What people think of you 0.41 0.64 - The joy and satisfaction you get from life - The closeness between you and your partner 0.65 - The closeness between you and your parents 0.55 0.63 - Certainty in your life If you were to have a child in the next three years, to what extent would the following persons agree with your choice? 0.62 - Most of your friends - Your mother 0.78 0.71 - Your father The decision about whether to have children can depend on various situations. How much could your decision about whether to have a child in the next three years depend on. . . 0.72 - Your economic situation 0.67 - Your job - Your housing conditions 0.68 - Your health 0.56 0.70 - Your partner’s job - Help from non-cohab. relatives in caring for the children 0.65 - Help from your partner in caring for the children 0.69 Note: The loadings shown are those that are useful for placing the item in the factor (>.04). Table A2 Variables considered in the analysis together with descriptive statistics (N = 2871). Var. name Block a: background variables Number of children NCh AgeW Age of woman CDur Couple’s duration CTy Type of couple Reg Region of residence MunS Municipality size CEd Couple’s levels of education WE Woman’s employment situation Modalities FREQ (%) 0 (Ref.) 1 2+ <30 30–40 (Ref.) >40 0–4 5–10 years (Ref.) >10 Married (Ref.) Cohabiting North (Ref.) Center South - Islands Big (Ref.) Medium Small Both low (Ref.) Both medium Both high Her > Him Him > Her Public sector (Ref.) Priv. sect./perm. contr. Priv. sect./temp. contr. Not working 12.7 30.3 57.1 9.5 54.2 36.3 15.5 20.9 63.6 96.2 3.8 48.9 17.0 34.1 16.8 39.6 43.6 24.1 29.7 6.1 22.9 17.2 21.0 35.5 4.8 38.8 25 L. Mencarini et al. / Advances in Life Course Research 23 (2015) 14–28 Table A2 (Continued ) Var. name ME Man’s employment situation CHD Current housework division SHD Woman’s satisfaction with housework division Rel Religiosity Sib Siblings Block b: primary antecedents of fertility intentions PAtt Positive attitudes NAtt Negative attitudes SubN Subjective norms PBC Perceived behavioral control Block c: fertility intentions Fint Fertility intentions Block d: constraints 2003–2007 Couple’s disruption CDis Block e: fertility outcomes Birth of a child Child Modalities FREQ (%) Public sector (Ref.) Priv. sect./perm. contr. Priv. sect./temp. contr. Not working <95% women (Ref.) 95% women Yes/moderate (Ref.) Not at all At least once per month (Ref.) Rarely/never Both partners without siblings At least one partner with two or more children Other (Ref.) 19.9 73.8 2.9 3.3 65.6 34.4 4.3 95.7 55.2 44.8 2.4 29.9 67.7 Class Class Class Class Class Class Class Class Class Class Class Class 1 2 3 1 2 3 1 2 3 1 2 3 8.8 79.7 11.5 5.4 86.5 8.1 12.0 71.6 16.4 13.9 70.8 15.3 Definitely not Probably not Probably yes Definitely yes 48.7 24.7 16.2 10.5 No (Ref.) Yes 96.1 3.9 No (Ref.) Yes 87.2 12.8 Table A3 Detailed model results. Stepwise procedure for model selection with comparison of the reduced model to the full model by means of the Likelihood Ratio test, to account for the multiple test problem (significance level set at 0.05). Variables Categories Coeff. Block b: primary antecedents of fertility intentions Cumulative logit model predicting ‘‘positive attitudes’’ toward childbearing 0 (Ref.) Number of children 1 2+ Couple’s duration 0–4 5–9 years (Ref.) 10 At least once per month (Ref.) Religiosity Rarely/never Class 1 Negative attitudes Class 2 Class 3 (Ref.) Subjective norms Class 1 Class 2 Class 3 (Ref.) 0 0.205 0.590 0.265 0 0.671 0 0.228 1.471 2.983 0 0.792 2.426 0 Cumulative logit model predicting ‘‘negative attitudes’’ toward childbearing Number of children 0 (Ref.) 1 2+ Region of residence North (Ref.) Center South 0 0.198 0.408 0 0.091 0.323 St. errors t-Stat 0.161 0.178 0.162 1.274 3.312 1.636 0.158 4.242 0.099 0.197 0.253 2.306 7.461 11.766 0.153 0.205 5.171 11.852 0.198 0.200 0.998 2.038 0.156 0.129 0.584 2.496 26 L. Mencarini et al. / Advances in Life Course Research 23 (2015) 14–28 Table A3 (Continued ) Variables Categories Current housework division <95% women (Ref.) 95% women Class 1 Class 2 Class 3 (Ref.) Class 1 Class 2 Class 3 (Ref.) Class 1 Class 2 Class 3 (Ref.) Positive attitudes Subjective norms Pbc Cumulative logit model predicting ‘‘subjective norms’’ toward childbearing 0 (Ref.) Number of children 1 2+ <30 Age of woman 30–40 (Ref.) >40 Couple’s duration 0–4 5–9 years (Ref.) 10 Region of residence North (Ref.) Center South Both low (Ref.) Couple’s levels of education Both medium Both high Her > Him Him > Her Siblings Both partners without siblings At least one partner with large family Other (Ref.) Class 1 Positive attitudes Class 2 Class 3 (Ref.) Negative attitudes Class 1 Class 2 Class 3 (Ref.) Coeff. 0 0.271 1.623 2.906 0 0.713 0.783 0 0.189 0.845 0 0 1.194 2.278 0.189 0 0.391 0.159 0 0.686 0 0.435 0.004 0 0.155 0.609 0.069 0.100 0.631 0.394 0 0.971 2.479 0 0.088 0.697 0 Cumulative logit model predicting the ‘‘perceived behavioral control’’ over childbearing 0 (Ref.) 0 Number of children 1 0.565 0.407 2+ Age of woman <30 0.351 30–40 (Ref.) 0 0.335 >40 Current housework division <95% women (Ref.) 0 95% women 0.272 Positive attitudes Class 1 0.582 0.194 Class 2 Class 3 (Ref.) 0 Negative attitudes Class 1 0.102 0.978 Class 2 Class 3 (Ref.) 0 Block c: fertility intentions Cumulative logit model predicting fertility intentions Number of children 0 (Ref.) 1 2+ <30 Age of woman 30–40 (Ref.) >40 Couple’s duration 0–4 5–9 years (Ref.) 10 Public sector (Ref.) Woman’s employment situation Priv. sect./perm. contr. Priv. sect./temp. contr. Not working 0 0.961 2.195 0.278 0 1.125 0.379 0 1.144 0 0.118 0.502 0.218 St. errors t-Stat 0.122 0.166 0.250 2.230 9.768 11.644 0.165 0.238 4.311 3.289 0.169 0.209 1.115 4.048 0.143 0.161 0.178 8.330 14.193 1.061 0.106 0.163 3.688 0.979 0.172 3.990 0.124 0.102 3.506 0.038 0.124 0.203 0.132 0.141 0.301 0.292 1.249 2.999 0.522 0.710 2.092 1.347 0.155 0.208 6.286 11.940 0.198 0.249 0.445 2.795 0.141 0.137 0.147 4.016 2.958 2.386 0.089 3.753 0.087 0.149 0.193 3.125 3.912 1.004 0.187 0.235 0.546 4.164 0.130 0.143 0.149 7.400 15.384 1.857 0.100 0.139 11.301 2.726 0.147 7.786 0.109 0.202 0.109 1.089 2.487 1.989 27 L. Mencarini et al. / Advances in Life Course Research 23 (2015) 14–28 Table A3 (Continued ) Variables Categories Current housework division <95% women (Ref.) 95% women At least once per month (Ref.) Rarely/never Class 1 Class 2 Class 3 (Ref.) Class 1 Class 2 Class 3 (Ref.) Class 1 Class 2 Class 3 (Ref.) Class 1 Class 2 Class 3 (Ref.) Religiosity Positive attitudes Negative attitudes Subjective norms Pbc Block d: constraints between 2003 and 2007 Logit model predicting the disruption of couples’ relationships Age of woman <30 30–40 (Ref.) >40 Couple’s duration 0–4 5–9 years (Ref.) 10 Married (Ref.) Type of couple Cohabiting Municipality size Big (Ref.) Medium Small Woman’s satisfaction Yes/moderate (Ref.) Not at all Regarding housework division Block e: fertility outcome Logit model predicting having a child Number of children Age of woman Couple’s duration Woman’s satisfaction Regarding housework division Fertility intentions Couple’s disruption 0 (Ref.) 1 2+ <30 30–40 (Ref.) >40 0–4 5–9 years (Ref.) 10 Yes/moderate (Ref.) Not at all Definitely not (Ref.) Probably not Probably yes Definitely yes No (Ref.) Yes References Ajzen, I. (1991). The theory of planned behavior. Organizational Behavior and Human Decision Processes, 50(2), 179–211. Ajzen, I. (2005). Attitudes, personality, and behavior (2nd ed.). Maidenhead, UK: Open University Press. Ajzen, I. (2010). Fertility intentions: Theory of planned behavior. Paper presented at BBAW/Leopoldina-Conference on Theoretical Foundations for the Analysis of Fertility, Lausanne, October 14–16, 2010.. Ajzen, I. (2011). Reflections on Morgan and Bachrach’s critique. Vienna Yearbook of Population Research, 9, 63–70. Ajzen, I., & Fishbein, M. (1973). Attitudinal and normative variables as predictors of specific behaviors. Journal of Personality and Social Psychology, 27, 41–57. Ajzen, I., & Fishbein, M. (1980). Understanding attitudes and predicting social behavior. Englewood-Cliffs, NJ: Prentice-Hall. Ajzen, I., & Klobas, J. (2013). Fertility intentions: An approach based on the theory of planned behavior. Demographic Research, 29(8), 203–232. Coeff. 0 0.197 0 0.303 0.893 1.759 0 0.189 1.034 0 0.539 1.306 0 0.361 0.188 0 0.609 0 0.841 1.224 0 0.340 0 3.906 0 1.056 0.035 0 1.853 0 0.466 0.156 0.375 0 1.494 0.885 0 1.553 0 2.207 0 0.647 1.950 3.236 0 1.487 St. errors t-Stat 0.086 2.298 0.080 0.185 0.219 3.796 4.834 8.015 0.171 0.245 1.103 4.217 0.146 0.180 3.692 7.253 0.120 0.151 3.006 1.246 0.429 1.421 0.460 0.506 1.829 2.416 0.485 0.701 0.369 10.590 0.469 0.408 2.253 0.086 0.517 3.586 0.178 0.226 0.186 2.613 0.691 2.012 0.291 0.179 5.137 4.954 0.212 7.314 0.754 2.926 0.232 0.229 0.250 2.785 8.536 12.935 0.414 3.589 Barber, J. (2011). The Theory of Planned Behavior: Considering dives, proximity and dynamics. Vienna Yearbook of Population Research, 9, 31–35. Berrington, A. (2004). Perpetual postponers? Women’s, men’s and couples’ fertility intentions and subsequent fertility behavior. Population Trends, 9–19. Berrington, A., Hu, Y., Smith, P. W. F., & Sturgis, P. (2008). A graphical chain model for reciprocal relationships between women’s gender role attitudes and labour force participation. Journal of the Royal Statistical Society A, 171(1), 89–108. Berrington, A., & Pattaro, S. (2013). Educational differences in fertility desires, intentions and behaviour. A life course perspective. Advance in Life Course Research (on line first). Billari, F. C., Philipov, D., & Testa, M.-R. (2009). Attitudes, norms and perceived behavioral control: Explaining fertility intentions in Bulgaria. , 25(4), 439–465. Bongaarts, J. (1992). Do reproductive intentions matter? International Family Planning Perspective, 18(3), 102–108. Bongaarts, J. (2001). Fertility and reproductive preferences in post-transitional societies. Population and Development Review, 27, 260–281. 28 L. Mencarini et al. / Advances in Life Course Research 23 (2015) 14–28 Borgoni, R., Berrington, A. M., & Smith, P. W. F. (2012). Selecting and fitting graphical chain models to longitudinal data. Quality and Quantity, 46(3), 715–738. Bühler, C. (2008). On the structural value of children and its implication on untended fertility in Bulgaria. Demographic Research, 18, 569–610. Cavalli, L., & Klobas, J. (2013). How expected life and partner satisfaction affect women’s fertility outcomes: The role of uncertainty in intentions. Population Review, 52(2), 70–86. Cox, D. R. (2008). On an internal method for deriving a summary measure. Biometrika, 95, 1002–1005. Cox, D. R., & Wermuth, N. (1993). Linear dependences represented by chain graphs. Statistical Science, 8, 204–283. Cox, D. R., & Wermuth, N. (1996). Multivariate dependencies. Models, analysis and interpretation. London: Chapman and Hall. Dalla Zuanna, G. (2001). The banquet of Aeolus. An interpretation of Italian lowest low fertility. Demographic Research, 4(5), 1–35. De Rose, A., Racioppi, F., & Zanatta, A. L. (2008). Italy: Delayed adaptation of social institutions to changes in family behavior. Demographic Research, 19, 665–704. Dommermuth, L., Klobas, J., & Lappegård, T. (2011). Now or later? The Theory of Planned Behavior and timing of fertility intentions. Advances in Life Course Research, 16(1), 42–53. Dommermuth, L., Klobas, J., & Lappegård T. (2013). The theory of planned behavior and the realization of fertility intentions. Paper presented to the conference ‘‘Changing families and fertility choices’’, Oslo, 6–7 June 2013.. Edwards, D. (2000). Introduction to graphical modeling (second ed.). New York: Springer-Verlag. Fishbein, M., & Ajzen, I. (1975). Belief, attitude, intention, and behavior: An introduction to theory and research. Reading, MA: Addison-Wesley. Fishbein, M., & Ajzen, I. (2010). Predicting and changing behavior: The reasoned action approach. New York: Psychology Press. Frydenberg, M. (1990). The chain graph Markov property. Scandinavian Journal of Statistics, 17, 333–353. Goldstein, J., Lutz, W., & Testa, M. R. (2004). The emergence of sub-replacement family size ideals in Europe. Population Research and Policy Review, 22, 479–496. Gottard, A. (2007). On the inclusion of bivariate marked point processes in graphical models. Metrika, 66, 269–287. Heiland, F., Prskawetz, A., & Warren, C. S. (2008). Are individuals’ desired family size stable? Evidence from West German panel data. European Journal of Population, 24, 129–156. Jaccard, J., & Davidson, A. R. (1975). A comparison of two models of social behavior: Results of a survey sample. Sociometry, 48, 497–517. Jorgensen, S. R., & Adams, R. P. (1988). Predicting Mexican-American family planning intentions: An application and test of a social psychological model. Journal of Marriage and the Family, 50, 107–119. Kapitány, B., & Spéder, Z. (2012). Realization, postponement or abandonment of childbearing intentions in four European countries. Population, 67(4), 599–629. Klobas, J. (2011). The Theory of Planned Behavior as a model of reasoning about fertility decisions. Vienna Yearbook of Population Research, 9, 47–54. Kuhnt, A. K., & Trappe, H. (2013). Easier to say than done: Childbearing intentions and their realization in a short term perspective (MPIDR WP 2013-018). Lauritzen, S. L. (1996). Graphical models. Oxford: Oxford Science Publications. Lauritzen, S., & Wermuth, N. (1989). Graphical models for associations between variables, some of which are qualitative and some quantitative. 17, 31–57. Liefbroer, A. C. (2009). Changes in family size intentions across young adulthood: A life-course perspective. European Journal of Population, 25(4), 363–386. Liefbroer, A. C. (2011). On the usefulness of the Theory of Planned Behavior for fertility research. Vienna Yearbook of Population Research, 9, 55–62. Mencarini, L., & Tanturri, M. L. (2006). High fertility or childlessness: Microlevel determinants of reproductive behavior in Europe. Population, 4, 389–416. Miller, W. B. (2011). Comparing the TPB and the T-D-I-B framework. Vienna Yearbook of Population Research, 9, 19–29. Mills, M., Mencarini, L., Tanturri, M. L., & Begall, K. (2008). Gender equity and fertility intentions in Italy and the Netherlands. Demographic Research, 18, 1–26. Mohamed, W. N., Diamond, I., & Smith, P. W. F. (1998). The determinants of infant mortality in Malaysia: A graphical chain modelling approach. Journal of the Royal Statistical Society A, 161, 349–366. Morgan, S. P. (1982). Parity-specific fertility intentions and uncertainty: The United Stated, 1970 to 1976. Demography, 19(3), 315–334. Morgan, S. P., & Bachrach, C. A. (2011). Is the Theory of Planned Behavior an appropriate model for human fertility? Vienna Yearbook of Population Research, 9, 11–18. Neyer, G., Lappegard, T., & Vignoli, D. (2013). Gender equality and fertility: Which equality matters? European Journal of Population, 29, 245–272. Noack, T., & Østby, L. (1985). Fertility expectations: A short-cut or dead-end in predicting fertility? Scandinavian Population Studies, 7, 48–59. Noack, T., & Østby, L. (2002). Free to choose – But unable to stick it? Norwegian fertility expectations and subsequent behavior in the following 20 years. In E. Klijzing, & M. Corijn (Eds.), Dynamics of fertility and partnership in Europe – Insights and lessons from comparative research (Vol. II). New York/Geneva: United Nations. Philipov, D. (2009). Fertility intentions and outcomes: The role of policies to close the gap. European Journal of Population, 25, 355–361. Philipov, D. (2011). Theories on fertility intentions: A demographer’s perspective. Vienna Yearbook of Population Research, 9, 37–45. Philipov, D., Spéder, Z., & Billari, F. C. (2006). Soon, later, or ever? The impact of anomie and social capital on fertility intentions in Bulgaria (2002) and Hungary (2001). Population Studies, 60(3), 289–308. Quesnel-Vallée, A., & Morgan, S. P. (2003). Missing the target? Correspondence of fertility intentions and behavior in the U.S. Population Research and Policy Review, 22, 497–525. Régnier-Loilier, A. (2006). Influence of own siblings ship size on number of children desired at various times of life. The case of France. Population, 61(3), 165–194. Régnier-Loilier, A., & Vignoli, D. (2011). Fertility intentions and obstacles to their realization in France and Italy. Population-E, 66(2), 361–390. Rinesi, F., Pinnelli, A., Prati, S., Castagnaro, C., & Iaccarino, C. (2011). Avoir un deuxième enfant en Italie: de l’intention à la réalisation. Population-F, 66(2), 435–450. Schoen, R., Astone, N. M., Kim, Y. J., Nathanson, C. A., & Fields, J. M. (1999). Do fertility intentions affect fertility behavior? Journal of Marriage and the Family, 61(3), 790–799. Smith, P. W. F., Berrington, A., & Sturgis, P. (2009). A comparison of graphical models and structural equation models for the analysis of longitudinal survey data. In P. Lynn (Ed.), Methodology of longitudinal surveys (pp. 381–392). Chichester, UK: Wiley. Spéder, Z., & Kapitány, B. (2014). Failure to realize fertility intentions: A key aspect of the post-communist fertility transition. Population Research and Policy Review (on line first). Symeonidou, H. (2000). Expected and actual family size in Greece: 1983– 1997. European Journal of Population, 4, 335–352. Testa, M.-R, & Grilli, L. (2006). The influence of childbearing regional contexts on ideal family size in Europe. Population-E, 61(1–2), 109–137. Thomson, E. (1997). Couple childbearing desires, intentions, and births. Demography, 34(3), 343–354. Thomson, E., McDonald, E., & Bumpass, L. L. (1990). Fertility desires and fertility: Hers, his, and theirs. Demography, 27(4), 579–588. Toulemon, L., & Testa, M.-R. (2005). Fertility intentions and actual fertility: A complex relationship. Population et Societies, 415. Vignoli, D., Rinesi, F., & Mussino, E. (2013). A home to plan the first child? Fertility intentions and housing conditions in Italy. Population, Space, and Place, 19, 60–71. Vikat, A., Spéder, Z., Pailhé, A., Pinnelli, A., Solaz, A., Beets, G., et al. (2007). Generations and Gender Survey (GGS). Towards a better understanding of relationships and processes in the life course. Demographic Research, 17, 389–400. Wright, S. (1921). Correlation and causation. Journal of Agricultural Research, 20, 557–585.