Preconception Motivation and Pregnancy ... - Wiley Online Library

7 downloads 0 Views 549KB Size Report
Child Care, we tested a mother-child influence framework hypothesized to mediate between the mother's preconception motivation and preg- nancy wantedness ...
WARREN B. MILLER

Transnational Family Research Institute

MARJORIE R. SABLE

University of Missouri-Columbia*

JONATHON J. BECKMEYER

University of Missouri-Columbia**

Preconception Motivation and Pregnancy Wantedness: Pathways to Toddler Attachment Security

This research was designed to increase our understanding of how the motivational antecedents to childbearing and emotional responses to pregnancy affect the subsequent attachment bond of a toddler to his or her mother. Using a sample of 1,364 mothers and their newborns from the Study of Early Child Care, we tested a mother-child influence framework hypothesized to mediate between the mother’s preconception motivation and pregnancy wantedness and her child’s security of attachment at 24 months. Almost all elements of the mediational sequence were confirmed in a structural equation model. We observed 2 unmediated pathways to attachment security, 1 from preconception motivation and 1 from parenting stress during infancy, and suggest interpretations of these unexpected pathways.

Transnational Family Research Institute, Aptos, CA 95003-4626 ([email protected]). *School of Social Work, 730 Clark Hall, University of Missouri, Columbia MO, 65211. **Department of Human Development and Family Studies, 314 Gentry Hall, University of Missouri, Columbia, MO 65211. Key Words: attachment, depression, mother-child relations, parenthood, pregnancy, stress.

1174

According to social bonding theory (Miller & Rodgers, 2001), there is a biologically based, experientially shaped system that acts as a motivational substrate to promote bonding between humans, one component of which enables the nurturant bonding of parents to their children and another component of which enables the reciprocal succorant bonding of children to their parents. (Succorance is a term coined by Murray, 1938, to describe the general tendency to seek the help and protection of others.) The motivations and emotions that arise from the nurturant substrate are essential for successful reproduction because of the major impact they have on the decisions and behaviors that lead to pregnancy, the psychological and behavioral responses to a pregnancy once it occurs, and the patterns of parenting that follow the birth of a child (Miller, 2003). Because this component of the motivational substrate is so critical to the formation of maternal and paternal bonds, we explore here how its effects on feelings and behavior before, during, and after pregnancy influence the child’s attachment bond with the mother and father. In previous research, Miller, Feldman, and Pasta (2002) showed in a middle-class married sample that childbearing motivational traits measured months and years before a conception

Journal of Marriage and Family 71 (December 2009): 1174 – 1192

Pregnancy Wantedness and Toddler Attachment

1175

predicted the attachment security of 78 2to 4.5-year-old children resulting from that conception and that this effect was mediated by maternal warmth and child temperament. In a second study that involved a largely unmarried, Black sample of Early Head Start applicants, Miller, Sable, and Csizmadia (2008) showed in a multivariate analysis that both positive and negative maternal motivations measured during the pregnancy, although not pregnancy wantedness itself, predicted attachment security in 78 14-month-old children. No appropriate variables were available in this case to study the mediators between maternal motivations and attachment security. In the current study, we reexamine important findings from both of these previous studies in a much larger sample, using hypotheses about how both preconception motivations and pregnancy wantedness predict toddlers’ security of attachment through a set of mediator variables.

promote human bonding in general and nurturant bonding in particular. Additional substrate traits relevant to reproduction are derived from behavioral domains other than that of nurturant bonding and serve to promote behaviors such as those related to work or educational activities that may be antithetical to nurturant bonding. Here we hypothesize that the motivational substrate for childbearing influences the mother in a sequence, first on whether or not she becomes pregnant, then on her emotional response to the pregnancy, and then on her responses to the child after birth, including her parenting behavior and interaction with the child. In the course of this sequence, we hypothesize that the child’s attachment to the mother is shaped. We call this approach the motherchild influence framework and emphasize that it represents the effect that the substrate has directly on the mother’s motivations and emotions, and indirectly through those motive forces, on her decisions and behaviors across time. In Figure 1 we depict the mother-child influence framework in terms of the theoretical constructs and time frame that were available within the data base used in the research reported here, the Study of Early Child Care (SECC). The figure is organized in two dimensions. Moving from top to bottom, as indicated in the left-hand column, it is organized according to the TDIB constructs. Intentions are measured only as part of preconception motivations (see Method). Several states—depression and stress—are secondary to motivational states (see the next paragraph). Moving from left to right, as indicated in the two top rows, the figure is organized across time, beginning before pregnancy and ending with toddlerhood at 24 months. Four steps in the motivational-behavioral sequence are represented in Figure 1. The first step involves the two variables of preconception motivation and pregnancy wantedness. Both of these are state variables. Preconception motivation reflects primarily the woman respondent’s preconception desires and intentions. Pregnancy wantedness reflects her feelings in response to being pregnant. Although previous research (Miller, 1995) has shown explicit childbearing motivational traits (i.e., those that are available to consciousness and are typically measured with self-report techniques) to be very good predictors of state measures of motivation such as

THEORETICAL FRAMEWORK To develop our hypotheses, we drew upon a theoretical framework called the Traits-DesiresIntentions-Behavior (TDIB) framework that was initially designed by Miller (1994) to explain how motivational traits influence the occurrence of fertility events. The three basic ideas of this framework are that the traits that motivate humans toward or away from childbearing are experienced as conscious desires for or against having a child, that these desires are typically transformed into conscious intentions to have or avoid having a child through some decisionmaking process, and that these intentions are then implemented through specific behaviors that are designed to achieve or avoid pregnancy and the bearing of a child. The motivational traits that are activated in the form of conscious desires for or against childbearing constitute an unconscious motivational substrate. These traits may be thought of as enduring dispositions of the human organism to respond positively or negatively to infants and children and the nurturant feelings that their behavior tends to elicit. These traits are biologically based (Miller & Rodgers, 2001) and experientially shaped (Miller, 1992) and play an important role in both planned and unplanned pregnancies (Miller, 2003; Miller & Pasta, 2002). Many of the traits that constitute the motivational substrate for childbearing serve to

1176

Journal of Marriage and Family FIGURE 1. THEORETICAL FRAMEWORK. Time Before Conception

During Pregnancy

At 1 Month

At 6 Months

At 15 Months

At 24 Months

Maternal/Child Influence Framework

Maternal Traits

Maternal States

Maternal Personality

Preconception Motivation Pregnancy Wantedness

Maternal Depression

Infancy

Maternal Behavior

Maternal/Child Behavior

Child Traits

Parenting Stress Toddlerhood

Parenting Behavior

Mother/Child Interaction

Child Attachment

Note: Figure shows how the constructs of the mother-child influence framework (on left, from top to bottom) are distributed across time (at top, from left to right), depicting the overall pathway through which maternal motivations that begin before pregnancy influence the formation of child attachment at age 2. Horizontal lines indicate the time period covered by the constructs measured and used in this study.

desires and intentions (Miller, 1994), no motivational trait variables were available in the SECC data. Other research has shown that certain maternal personality traits are good predictors of explicit childbearing motivations (Miller, Pasta, MacMurray, Muhleman, & Comings, 2000) and therefore could serve as proxies for childbearing motivational trait in our modeling efforts. Several personality traits measures were available in the data, including two (Agreeableness and Extraversion) that have been shown to be good

predictors of positive childbearing motivations and one (Neuroticism) that may be predictive of negative childbearing motivation (Miller, 1992; Miller et al., 2000). Therefore we hypothesize that at least one and possibly all three will predict preconception motivational states. On the basis of our previous research (Miller, Sable, & Csizmadia, 2008; Miller & Jones, 2008), we hypothesize that preconception desires and intentions will be strong predictors of pregnancy wantedness and that one or more personality

Pregnancy Wantedness and Toddler Attachment

1177

traits will have an effect on this wantedness outcome as well. In addition to testing our TDIB motivational framework, the inclusion of the trait variables in the structural equation model adds stability to the preconception motivation and pregnancy wantedness variables. The next three steps in the motivationalbehavioral sequence involve variables that we hypothesize act as mediators of the relationship between preconception motivation and pregnancy wantedness on one hand and child attachment security on the other. The first mediational step involves the prediction of maternal depression and parenting stress by pregnancy wantedness. Miller and Pasta (2002) have shown negative motivations about the responsibilities of child care to be a strong predictor of when a pregnancy is not wanted. In terms of social bonding theory, pregnancy and caring for an infant represent energetic and time demands on the woman. When she is not motivationally prepared for these challenges, we hypothesize that there will be a serious risk of depression or a stress response or both. Recent literature supports this conclusion. Pregnancies that are not wanted are associated with maternal depression during pregnancy (Kitamura, Sugawara, Sugawara, Toda, & Shima, 1996; Rich-Edwards et al., 2006), during the postnatal period (Beck, 2001; Csatordai et al., 2007; Rich-Edwards et al.), and for an extended period of mothers’ reproductive careers (Barber, Axinn, & Thorton, 1999). Mothers who have experienced depression during the postnatal period report higher levels of parenting stress (Cornish et al., 2006), in some cases for 3 to 4 years (Milgrom, Ericksen, McCarthy, & Gemmill, 2006). A less than fully wanted pregnancy may also contribute to parenting stress during the toddler years (Ispa, Sable, Porter, & Csizmadia, 2007). Such an extension of an unwanted pregnancy’s effects on parenting stress beyond infancy into toddlerhood leads us to distinguish these two periods in our theoretical framework. The second mediational step involves the prediction of parenting behavior and mother-child interaction by maternal depression and parenting stress. In terms of social bonding theory, maternal depression and its associated affective withdrawal, decreased energy, and self-focus would be expected to reduce the attention and effort that can be invested in the bond with the child. Belsky, Steinberg, and Draper (1991),

who summarized the earlier relevant literature in developing their evolutionary theory of socialization, concluded that under conditions of stress ‘‘parental behavior becomes less affectively positive and more insensitive, perhaps to the point of . . . abuse’’ (p. 654). Simpson’s (1999) observations that the effects of economic, occupational, marital, and psychological stress on families tend to reduce the quality of parenting reinforced these conclusions. Additional research supports the relationships hypothesized here between maternal depression and stress on one hand and parenting behavior on the other (Buist, 1998; Lovejoy, Graczik, O’Hare, & Neuman, 2000). More recent studies reinforce these earlier findings for postnatal depression (Albright & Tamis-LeMonda, 2002; Coyne, Low, Miller, Seifer, & Dickstein, 2007; Pelaez, Field, Pickens, & Hart, 2008), for parenting stress (Chang et al., 2004; Diener, Nievar, & Wright, 2003; Tarabulsy et al., 2008), for both depression and stress (Anhalt, Telzrow, & Brown, 2007), and for depression as mediated by stress (Gerdes et al., 2007). The third mediational step involves the prediction of attachment security by parenting behavior—especially maternal warmth—and mother-child interaction—especially maternal sensitivity. The latter is, of course, the link originally proposed by Ainsworth (1973) as an extension of Bowlby’s theory of attachment. This link has subsequently been supported by extensive research (Belsky, 1999), although the meta-analysis of De Wolff and IJzendoorn (1997) indicated that maternal sensitivity was only modestly associated with child security of attachment. This has prompted us to look beyond maternal sensitivity for additional maternal characteristics in our data. Recent work has suggested what these characteristics might be by establishing the importance of emotional availability for mother-child relationships (Biringen, 2000), demonstrating the role of ‘‘mind-mindedness’’ in maternal sensitivity (Laranho, Bernier, & Meins, 2008) and showing that maternal responsiveness facilitates infants’ social, emotional, communicative, and cognitive development more generally (Landry, Smith, & Swank, 2006). There are potential connections between earlier and later steps in the mediational sequence that may bypass the intermediate steps that we have hypothesized. For example, Carter, Garrity-Rokous, Chazan-Cohen, Little, and

1178

Journal of Marriage and Family

Briggs-Gowan (2001) have shown that maternal depression may directly predict poor motherinfant interactions. Rather than proposing hypotheses about predictions that bypass steps in the mediational sequence, however, we allow our modeling of the four steps to indicate where such bypasses occur. The sample used in this study is not a nationally representative one but rather was recruited from hospitals at 10 collection sites. Because there is a strong possibility of demographic differences between sites and a reasonable possibility that these differences are significantly associated with some variables used in our model, we controlled for the major maternal and child demographic variables available in the SECC data base, including maternal age, education, marital status, and income and the child’s gender and birth order. Doing this also helped to control for sample attrition, which (as we discuss below) was significantly associated with the selected demographic variables. METHOD Design The data used to test our theoretical framework were collected during the Study of Early Child Care, a large-scale, longitudinal study of the effects of early child-care arrangements on children’s development, sponsored by the National Institute of Child Health and Human Development (NICHD Early Child Care Network, 1994). Participants in the study were recruited during the first 11 months of 1991 from hospitals located in or near 10 metropolitan areas in all sections of the United States. A total of 1,364 families with full-term healthy newborns were enrolled. Participants were selected on the basis of a conditionally random sampling plan that was designed to ensure that (a) recruited families included both mothers who planned to work or go to school part time or full time during the child’s first year as well as those who planned not to work or go to school and (b) recruited families reflected the economic, educational, and ethnic diversity of the 10 sites. Exclusionary criteria included mothers younger than 18, families not planning to remain in the catchment area, newborns with disabilities or serious illness, and mothers not conversant in English. Phase I data collection occurred when the children were 1,

6, 15, 24, and 36 months of age and involved a variety of telephone, in-home, and laboratory interview and observational techniques (fully documented at http://secc.rti.org). The current study uses data through the first four collection periods of phase I (24 months), at which time sample size had declined by 162 (11.9%) to 1,202. Participants We included all 1,364 initial participants in our analyses, although loss to follow-up and the noncompletion of some instruments resulted in somewhat smaller numbers of participants at different data collection points for different measures. The different ns for all pairs of model and control variables that were created by attrition and instrument noncompletion are reported below the matrix diagonal in the Appendix. Table 1 shows the mother and child characteristics of the initial participants. Maternal age, education, marital status (recoded 1 = married, 2 = not married), family income, and child gender and birth order were all used as control variables in data analysis. We conducted t tests to determine the effect of sample attrition between the 1-month and 24month data collection points on the control variables. All four of the maternal variables were significantly affected by attrition, with t values ranging from −2.82(p = .005) to −4.91(p < .001). The negative scores indicate that those missing at the 2-year follow-up have a lower score on the variable. Neither of the child variables was significantly affected by attrition. We further discuss our handling of missing data in the Measures and Data Analysis sections below. Measures We used variables representing each of the eight mother-child influence framework categories shown in Figure 1 to test the proposed theoretical framework. We discuss these variables in turn, beginning with those representing maternal personality and ending with the variable representing child attachment security. For all variables and each of their components we report means and standard deviations in parentheses. In several cases, we also report a variable’s minimum and maximum scores. Where appropriate, we separately report a variable’s coefficient. Any variable with at least one component

Pregnancy Wantedness and Toddler Attachment

1179

Table 1. Mother and Child Characteristics (N = 1,364 Unless Otherwise Indicated) Variable

Mean

SD

Maternal age Maternal education, years Maternal marital status Married Not married, cohabiting Not married, not cohabiting Maternal ethnicity African American European American Other Maternal employment status Employed, at work Employed, on leave Not employed Annual family income, thousand dollars (n = 1,273) Child gender Male Female Child birth order First Second Third Fourth Fifth and above

28.11 14.23

5.63 2.51

missing was declared missing. We conducted t tests to determine the effect of sample attrition between the 1-month and 24-month data collection points on the predictor variables. Both preconception motivation (t = −3.69, p < .001) and pregnancy wantedness (t = −2.42, p < .001) were significantly affected by attrition. Again, the negative scores indicate that those missing at the 2-year follow-up have a lower score on the variable. Maternal personality. The SECC used the NEO Personality Inventory and the NEO Five-Factor Inventory (Costa & McCrae, 1992) to measure three maternal personality constructs at the 1-month data collection point. Because the personality constructs were designed as trait measures, we assume in our modeling that the obtained scores closely reflect the respondents’ trait status prior to the pregnancy. Neuroticism (M = 29.8, SD = 7.16) was based on only three facets of the original six-facet NEO PI scale, including Anxiety, Hostility, and Depression. Extraversion (M = 42.5, SD = 5.83) was based on only two facets of the original six-facet

%

76.5 8.9 14.5 12.8 82.6 4.6 9.9 50.8 39.3 37.9

34.1 51.7 8.3 44.8 34.8 14.6 4.5 1.3

NEO PI scale, including Warmth and Positive Emotions. Agreeableness (M = 46.3, SD = 5.29) was based on 12 items taken from the original six-facet NEO FFI scale. Preconception motivation and pregnancy wantedness. To measure these constructs, we used data from the SECC questionnaire called Your Pregnancy. This instrument was administered at the 1-month data collection point, asking retrospective questions about the preconception expectations of the woman and her partner as well as her feelings about the pregnancy when she found out she was pregnant, when the baby was born, and at the 1-month postnatal interview. Table 2 shows the five items of this questionnaire and their response frequencies. Table 3 shows the intercorrelations between these five items. For purposes of data reduction, we also examined the correlations of these five items with three variables used in testing our theoretical framework. Table 3 shows these correlations as well. The five item intercorrelations suggest the presence of two overlapping clusters, one that includes Items 1, 2, and 3 and one that

1180

Journal of Marriage and Family Table 2. Frequencies for the Five Items From the Your Pregnancy Questionnaire (N = 1,364)

Your Pregnancy Item

Frequency

1. When you got pregnant with your baby, were you expecting or trying to become pregnant? No Yes 2. Was the baby’s father expecting or hoping that you would become pregnant? No Yes 3. When you found out you were pregnant, how did you feel? 1. Very unhappy 2. Sort of unhappy 3. Mixed happy and unhappy 4. Sort of happy 5. Very happy 4. By the time the baby was born, how did you feel about having a baby? 1. Very unhappy 2. Sort of unhappy 3. Mixed happy and unhappy 4. Sort of happy 5. Very happy 5. How do you feel now about having a baby? 1. Very unhappy 2. Sort of unhappy 3. Mixed happy and unhappy 4. Sort of happy 5. Very happy

%

617 746

45.3 54.7

523 837

38.5 61.5

78 57 311 156 761

5.7 4.2 22.8 11.4 55.8

28 6 58 101 1,171

2.1 0.4 4.3 7.4 85.9

30 5 20 60 1,249

2.2 0.4 1.5 4.4 91.6

Table 3. Correlations of the Five Your Pregnancy Items With Themselves and With Three Selected Outcome Variables Used in Testing the Theoretical Framework Variable 1. Trying to become pregnant 2. Father hoping for pregnancy 3. When found out, how did you feel? 4. When baby born, how did you feel? 5. Now (baby at one month) how do you feel? 6. Maternal Depression, 15 months 7. Parenting Stress, 1 month 8. Security of Attachment, 24 months

1

2

3

4

5

.77∗∗∗ .61∗∗∗ .09∗∗∗ .06∗ −.14∗∗∗ −.03 .11∗∗∗

.52∗∗∗ .06∗ .03 −.11∗∗∗ −.03 .09∗∗

.35∗∗∗ .32∗∗∗ −.16∗∗∗ −.17∗∗∗ .13∗∗∗

.74∗∗∗ −.08∗∗ −.13∗∗∗ .06∗

−.03 −14∗∗∗ .02

Note: ns for correlations of Your Pregnancy items with each other and with Parenting Stress, 1 month = 1,358 – 1,364, for the Your Pregnancy items with Maternal Depression, 15 months = 1,238 – 1,241, and for the Your Pregnancy items with Secure Attachment, 24 months = 1,194 – 1,197. ∗ p < .05, **p < .01, ***p < .001.

includes Items 3, 4, and 5. Correlations of the five items with the three outcome variables show a differential correlation of these two clusters, 1 and 2, with different outcomes. On the basis of this evidence of internal consistency and differential outcome prediction, together with the meanings of the five items, we decided to

construct two variables. Because of the different response range for Items 1 and 2 relative to Items 3, 4, and 5, we used a principal components analysis to derive the weightings of items within each of the two separate clusters. In the first variable, which was called Preconception Motivation (M = 6.12, SD = 1.62, and

Pregnancy Wantedness and Toddler Attachment

1181

min/max scores = 2.61/7.62), Items 1, 2, and 3 were weighted .919, .884, and .801, respectively. In the second variable, which we called Pregnancy Wantedness (M = 11.1, SD = 1.61, and min/max scores = 2.40/12.00), Items 3, 4, and 5 were weighted .614, .897, and .889, respectively. We interpret these two clusters and the presence of Item 3 in both of them as follows. In the first cluster, Item 3 is given almost equal weight with the two items that reflect the respondents’ intentions and their perceptions of their partners’ intentions and desires (the latter measured as ‘‘hopes’’). No question was asked about the respondents’ desires, but we interpret the ‘‘feelings’’ when they found out they were pregnant in Item 3 as a good proxy for their preconception desires. This bipolar variable ranges between positive and negative motivation and is moderately skewed in the negative direction. In the second cluster, Item 3 is not equally weighted with the two items that reflect the respondents’ feelings about the pregnancy at birth and one month later. The frequencies for these two items indicate that many of the women with unhappy or mixed feelings have become very happy, leaving a small, core group of women standing at the negative pole of this variable and representing women who were least able over time to come around to wanting the pregnancy. This bipolar variable also ranges between positive and negative motivation but is less balanced, being more strongly skewed in the negative direction.

Parenting stress. The SECC used two different instruments to measure parenting stress. At 1 and 6 months parents completed a 30-item version of Abidin’s (1983) Parenting Stress Index (PSI), designed to identify parent-child systems under stress. Specifically, the SECC version summed three parental subscales that measured the mother’s sense of competence as a parent, her investment in the role of being a parent, and the degree of restriction she felt by the parental role. We summed the PSI scores at the two points in time to create a measure called Parenting Stress, Infancy (M = 103.4, SD = 18.8). Coefficient α for this variable was .801. At 15 and 24 months parents completed the Parent-Role Quality Scale or PRQS (Barnett & Marshall, 1991) for which participants are asked to rate 10 rewarding and 10 concerning parts of their experiences as parents. The reward score is subtracted from the concern score to arrive at a measure of parenting stress. We summed the PRQS scores at the two points in time to create a measure called Parenting Stress, Toddlerhood (M = 68.5, SD = 11.5). Coefficient α was .772.

Maternal depression. For the measurement of maternal depression, the SECC used the Center for Epidemiological Studies Depression Scale (CES-D; Radloff, 1977) at all four data collection points covered in the current study. Our measure of Maternal Depression (29.1, 21.0) consisted of the sum of the three CES-Ds scores obtained at 1, 6, and 15 months. Omitting from our depression measure a score obtained at 24 months, which was the same time that attachment security was measured, strengthened the predictive aspect of our model. (To achieve our goal of increasing variable reliability by averaging measures across at least two data collection points, we did include the 24-month measures in several subsequently described variables.) Coefficient α for this variable was .761.

Parenting behavior. The HOME Inventory (Caldwell & Bradley, 1984) was used by the SECC to measure the quality and quantity of support available to the child in the home environment, as evaluated by an observer during a home visit. The infant/toddler version was administered twice during the period covered by the current study. We summed the positive involvement subscale at 6 months (Items 1, 2, 5, 9, 10, 35) and the positive parenting subscale at 15 months (Items 1, 2, 5, 9, 10, and 11) to create a measure called Positive Parenting (M = 12.7, SD = 2.11). Coefficient α for this variable was .327. This value indicates a relatively low consistency from 9 to 15 months between the two components. Each component had a higher α than their composite, .516 and .555, respectively. Accordingly, we explored using two separate Positive Parenting variables within the model. The two-component composite variable performed more satisfactorily within the model (see Results), so we retained it, drawing two conclusions. First, we concluded that a big factor in the low reliability was that parenting behavior, much more than parenting states as reflected in the Maternal Depression and two Parenting Stress variables, changes considerably from when the child is an infant to when it is a toddler because of the very different

1182 demands of the child’s developing behavior. Second, we concluded that the two-component variable’s performance in the model suggested that there was sufficient commonality in the measure across the two time periods to justify its retention. Mother-child interaction. The SECC developed its own scales of mother-child interaction that were based on the previous work of other researchers concerned with the qualities of parenting that were important for the development of secure attachment. Motherchild interactions were rated from videotapes of free play between mother and child. Ratings scales focused on maternal sensitivity, intrusiveness, detachment, and stimulation of cognitive development and on the child’s mood and engagement with the mother. We examined the intercorrelations between scales measured at 15 and at 24 months and selected three at each time point with strong internal consistency, including sensitivity composite, stimulation of cognitive development, and child’s engagement with mother (average scale intercorrelation = .44) at 15 months and sensitivity to nondistress, stimulation of development, and engagement with mother (average scale intercorrelation = .49) at 24 months. Because the sensitivity composite had a threefold greater response range compared to the other scales, in order not to give it undue weight in the composite score, we adjusted its response range to be equivalent to the other scales. We next summed the three scales at each time point and then across the two time points to create a measure called MotherChild Engagement (M = 16.9, SD = 2.7). Its coefficient α was .748. Security of attachment. The SECC used Waters and Deane’s (1985) Attachment Q-Set to measure the child’s security of attachment at 24 months. This instrument is based on 90 items that are descriptive of a child’s current behavior. Representative items for attachment security include ‘‘Readily shares if mother asks’’ (high score is secure) and ‘‘Wants to be the center of mother’s attention’’ (low score is secure). During a 2-hr home visit a trained observer sorts 90 cards into nine separate piles on the basis of a predetermined distribution (4-6-10-15-20-1510-6-4). Each card contains an item description of a child’s behavior. Cards most characteristic of the index child are placed at one end of the

Journal of Marriage and Family distribution (piles 7, 8, and 9) and those least characteristic of the index child are placed at the opposite end (piles 3, 2, and 1). The resulting profile of the index child is then correlated with the profile of a prototypically secure child as measured by experts in the field to give an individual score (M = 0.291, SD = 0.206). The reliability and validity of this instrument is good (see http://secc.rti.org). The large number of items, nine-category response scale, and normal curve-based sorting procedure mean that this instrument has excellent distributional properties for use with correlational types of analysis. Data Analysis We based our analysis on the causal model that is implicit in the mother-child influence framework as represented by the theoretical constructs and time sequence shown in Figure 1. More explicitly, this framework may be viewed as a set of eight simultaneous regression analyses organized as follows, with the letters in parentheses indicating the 18 major hypotheses that emerge from our framework: Preconception Motivation is predicted by one or more of the three maternal personality variables (a); Pregnancy Wantedness is predicted by one or more of the three maternal personality variables (b) and by Preconception Motivation (c); Maternal Depression is predicted by Pregnancy Wantedness (d); Parenting Stress, Infancy is predicted by Pregnancy Wantedness (e) and Maternal Depression (f); Parenting Stress, Toddlerhood is predicted by Pregnancy Wantedness (g), Maternal Depression (h), and Parenting Stress, Infancy (i); Positive Parenting is predicted by Maternal Depression (j), Parenting Stress, Infancy (k), and Parenting Stress, Toddlerhood (l); Mother-Child Engagement is predicted by Maternal Depression (m), Parenting Stress, Infancy (n), Parenting Stress, Toddlerhood (o), and Positive Parenting (p); and Security of Attachment is predicted by Positive Parenting (q) and Mother-Child Engagement (r). In addition, for control purposes, we added as predictors in each of these regressions the four maternal and two child demographic variables discussed in connection with Table 1. Before constructing the structural equation model (SEM) in which all of the regressions were tested simultaneously, we conducted preliminary analyses in which each regression was run separately, using backward stepping elimination. We ran each regression first with

Pregnancy Wantedness and Toddler Attachment

1183

the number of cases available at the time point of the dependent variable, using listwise deletion of missing cases. We then repeated each regression with the number of cases restricted to those available at the 24-month follow-up, again using listwise deletion of missing cases. The only differences between each pair of unrestricted and restricted regressions was the loss in three restricted regressions of a single predictor that was significant at p < .10 but >.05 with the unrestricted n. For this reason we used only the p < .05 criterion when conducting our SEM analysis. We estimated the SEM according to the explicit framework described above, using LISREL software (Joreskog & Sorbom, 1996) and a covariance matrix of the 17 variables included in the regression analyses. We decided that the imputation of missing data in the covariance matrix was unnecessary for three reasons: The sensitivity analyses conducted with the preliminary regressions showed the elements of our model to be robust to the amount of data that was missing, the correlation matrix of the 17 variables was positive definite, and the sample size was very large. We therefore used the pairwise missing option available in LISREL. This had the added benefit of allowing us to use all available data for every pair of the 17 variables. The correlation matrix used to calculate the covariance matrix for LISREL and the pairwise ns upon which those correlations were based are given in the Appendix. For purposes of SEM estimation, we treated the eight dependent variables in the regression analyses (Preconception Motivation through Security of Attachment as shown in Figure 1) as endogenous variables (etas in LISREL terminology), locating them in the beta matrix. Because we were not interested in modeling their interrelationships, we treated the three maternal personality variables and six demographic control variables as exogenous variables (ksis in LISREL), locating them in the gamma matrix, and allowed them to freely correlate among themselves. Predictive connections among the endogenous variables and between the first two endogenous variables (Preconception Motivation and Pregnancy Wantedness) and the three maternal personality variables were entered in the model according to the explicit framework described above and an initial model was estimated. We then used modification indices (MIs)

to add additional predictive connections iteratively, beginning with the largest MI, and continuing until all connections with p < .05 were in the model. We then iteratively dropped all connections with p > .05, beginning with the least significant, and added back any connections that became significant during this process. Results of the modeling were not materially changed whether we based the calculation of the adding/dropping criteria on the maximum (1,364) or minimum (1,074) number of pairwise cases in the beta matrix. The model obtained through this process confirmed that Parenting Stress, Toddlerhood was predicted by Maternal Depression and Parenting Stress, Infancy, although not by Pregnancy Wantedness. In addition, it did not confirm that Parenting Stress, Toddlerhood predicted Positive Parenting, Mother-Child Interaction, or Security of Attachment. Therefore, even though this version of the model fit well, it contained a variable that did not pass through any effects of Preconception Motivation or Pregnancy Wantedness downstream to Security of Attachment. To simplify the model, we dropped Parenting Stress, Toddlerhood and reestimated the model. This had the added advantage of increasing the minimum n from 1,074 to 1,135 (see the Appendix). We note at this point that Parenting Stress, Infancy did pass through the effects of Preconception Motivation or Pregnancy Wantedness downstream to Security of Attachment. Because these two parenting ‘‘stress’’ variables are based on two distinct instruments with different approaches to measurement, we conclude that their different performance in the model reflects, at least in part, important differences in what is meant by the term ‘‘stress.’’ In one final analytic step, because of the low coefficient α of Positive Parenting, we reestimated the final model with Positive Parenting separated into the two Positive Parenting components, one at 6 months and the other at 15 months, and allowed a causal prediction from the first to the second of these components. Although this model’s overall fit was comparable to the overall fit of the final model version in which the two components were combined in one variable, the localized fit between Maternal Depression and the two causally related Positive Parenting components was relatively poor. We therefore rejected this version of our final model and report here that

1184

Journal of Marriage and Family

version in which the two components remained as one variable. RESULTS We present the results of the SEM analysis in two forms. Table 4 presents both the unstandardized and the standardized parameter estimates and t values for all the variables in the motherchild influence framework that significantly predict each outcome variable in the overall sequence. The R 2 is also given for each of the seven prediction equations, beginning with the final outcome variable, Security of Attachment, and moving backward to end with the initial outcome variable, Preconception Motivation. To help the reader visualize the hypothesized causal flow through the motherchild influence framework across time, Figure 2 presents the same information, showing the significant connections between all variables in the hypothesized model together with the standardized parameter estimate associated with each connection. Because we were especially interested in the differential effects

of Preconception Motivations and Pregnancy Wantedness on Security of Attachment we also estimated the standardized total and indirect effects of these two predictors on the child’s Security of Attachment score, which were 0.07/0.01 and 0.02/0.02, respectively. Not shown in either of these presentations are the significant connections of the six demographic control variables to the seven outcome variables that comprise the beta matrix. The standardized parameter estimates of these connections are shown in Table 5. The model results shown in Table 4, as adjusted for the control variables shown in Table 5, largely confirm the hypotheses derived from our framework. Of the 18 hypotheses, 13—including 2 related to the Parenting Stress, Toddlerhood variable that was dropped from the final model—were confirmed. When the five hypotheses that involved the dropped stress variable (numbers including g, h, i, l, and o) are excluded, of the 13 remaining hypotheses, 11 were confirmed, leaving only numbers k and l unconfirmed.

Table 4. Structural Equation Model, Showing Unstandardized Parameter Estimates (U.P.E.), Standardized Parameter Estimates (S.P.E.), and T Values for Prediction Pathways and R2 for Outcome Variables (N = 1,364) Outcome Variable Predictor Variables Security of Attachment Mother-Child Engagement Positive Parenting Parenting Stress, Infancy Preconception Motivation Mother-Child Engagement Positive Parenting Maternal Depression Positive Parenting Maternal Depression Parenting Stress, Infancy Maternal Depression Pregnancy Wantedness Maternal Depression Pregnancy Wantedness Pregnancy Wantedness Preconception Motivation Maternal Agreeableness Preconception Motivation Maternal Agreeableness ∗

p < .05, ***p < .001.

U.P.E.

S.P.E.

T Value

0.01 0.00 −0.00 0.01

0.19 0.06 −0.09 0.05

7.07*** 2.17* −3.59∗∗∗ 2.03∗

0.37 −0.01

0.19 −0.11

7.46∗∗∗ −4.59∗∗∗

−0.01

−0.09

−3.34∗∗∗

0.47 −1.49

0.52 −0.13

21.30∗∗∗ −5.36∗∗∗

−1.58

−0.12

−4.62∗∗∗

0.61 0.02

0.61 0.07

27.37∗∗∗ 3.32∗∗∗

0.02

0.07

2.57∗

R2 0.10

0.25

0.16 0.29

0.12 0.38

0.19

Pregnancy Wantedness and Toddler Attachment

1185

FIGURE 2. RESULTS OF THE LISREL STRUCTURAL EQUATION MODEL.

Time Before During Conception Pregnancy

At 1 Month

At 6 Months

At 15 Months

At 24 Months

Maternal/Child Influence Framework

Maternal Traits

Agreeableness +0.07

Maternal States

+0.07

Preconception Motivation +0.61

Pregnancy Wantedness

-0.12 Maternal Depression

-0.13

+0.52 Parenting Stress, Infancy

Maternal Behavior

-0.09 Positive Parenting

+0.06 Maternal/Child Behavior

-0.09

-0.11

+0.19

Mother/Child Engagement +0.19

Child Traits

+0.05

Security of Attachment

Note: Figure shows the pathways and their standardized parameter estimates connecting a maternal personality trait, preconception motivation, and pregnancy wantedness with subsequent maternal states and behavior, mother-child interaction, and child attachment. These findings are net the effects of maternal age, education, marital status, and income and the effects of child gender and birth order.

Because the three maternal personality variables were treated as exogenous variables in the gamma matrix, any nonhypothesized associations that they had with the beta matrix variables caused distortions in the overall fit of the model being tested. We therefore freed up those significant associations before calculating the overall model fit reported here. With 44 degrees of freedom, the chi-square of the model was 51.07 with p = .22. The root mean square error of approximation (RMSEA) was .011 and

the p value for a test of close fit (RMSEA