Maternal Accuracy and Behavior in Anticipating Children’s Responses to Novelty: Relations to Fearful Temperament and Implications for Anxiety Development

Abstract

Previous research has suggested that mothers’ behaviors may serve as a mechanism in the development from toddler fearful temperament to childhood anxiety. The current study examined the maternal characteristic of accuracy in predicting toddlers’ distress reactions to novelty in relation to temperament, parenting, and anxiety development. Ninety-three two-year-old toddlers and their mothers participated in the study. Maternal accuracy moderated the relation between fearful temperament and protective behavior, suggesting this bidirectional link may be more likely to occur when mothers are particularly attuned to their children’s fear responses. An exploratory moderated mediation analysis supported the mechanistic role of protective parenting in the relation between early fearful temperament and later anxiety. Mediation only occurred, however, when mothers displayed high accuracy. Results are discussed within the broader literature of parental influence on fearful children’s development.

Children’s early fearful temperament (i.e., behavioral inhibition; Kagan, Reznick, Clarke, Snidman, & Garcia-Coll, 1984) has been consistently linked to the development of anxiety and related problems (Biederman, Hirshfield-Becker, Rosenbaum, Herot, & Friedman, 2001; Schwartz, Snidman, & Kagan, 1999). Recent research has suggested that fearful temperament may elicit anxiogenic parenting (e.g., overprotection, intrusiveness), which then serves as a mechanism for future anxiety development. Anxiety prevention and intervention efforts would therefore benefit from better identifying mother-child dyads prone to this harmful mutual influence. Available evidence suggests that mothers’ perceptions of their children’s temperament predict parenting behaviors (Rubin, Nelson, Hastings, & Asendorpf, 1999) and strengthen the relation between temperament and parenting (Ghera, Hane, Malesa, & Fox, 2006). Relatively little is known, however, about the accuracy of mothers’ perceptions and how individual differences in this accuracy affect the link between fearful temperament and parent behaviors. We previously investigated this maternal characteristic of accuracy in anticipating children’s reactions to a variety of situations (Kiel & Buss, 2006). The current study focuses on maternal accuracy for novel situations, examining its effect on the association between fearful temperament and both protective and intrusive parenting behaviors. The current study also examined an exploratory model in which accuracy affects how maternal behavior mediates the development from fearful temperament to anxiety.

Parenting Behaviors Linked to Fearful Temperament and Anxiety

Parenting behaviors have been linked to both adaptive and maladaptive child development. Of particular relevance to the development of anxiety are parents’ interactions with their children during fear-provoking or potentially threatening situations. Kagan et al. (1984) reported that mothers of inhibited children described being distressed by shy/withdrawn behavior and often tried to socialize their children away from this behavioral style. Since then, more research has examined how protective and intrusive parenting behaviors relate to fearful temperament and early anxiety.

Protective Behaviors

Maternal behaviors that remove or shield children from a source of stress are often referred to as protective. Mothers may also show comforting and affectionate behaviors in response to children’s displays of fear, which may not remove the source but rather alleviates the stress associated with it. Both protective and comforting behaviors are often included in laboratory composites of maternal protection or similar parenting behavior (Rubin, Burgess, & Hastings, 2002; Rubin, Hastings, Stewart, Henderson, & Chen, 1997) because they prevent children’s independent coping. These behaviors have been linked to fearful temperament, anxiety, and the exacerbation of children’s fearfulness over time (Bayer, Sanson, & Hemphill, 2006; Chorpita & Barlow, 1998; Dadds & Roth, 2001; Hudson & Rapee, 2005; Kagan, Snidman, Arcus, & Reznick, 1994; Messer & Beidel, 1994; Nachmias, Gunnar, Mangelsdorf, Parritz, & Buss, 1996).

Protective parenting behaviors may be frequently linked to fearfulness and later anxiety because fearful children elicit them from their parents. So, not only do parents’ behaviors affect their children’s fear, but children’s fear also affects parents’ behaviors. Indeed, children’s shyness has been shown to predict parents’ lack of encouragement of independence (Rubin et al., 1999), and children’s anxious traits have been shown to predict parents’ warmth and protectiveness (Kendler, Sham, & MacLean, 1997), sometimes more strongly than parents’ characteristics (Rapee, 2001). Overprotective parenting also appears to occur specifically with anxious children rather than with all children of a particular parent (Hudson & Rapee, 2005), suggesting that the child’s anxiety plays a role in parenting. When children become fearful, they may express a desire to escape, avoid, or terminate the feared stimulus. It would certainly be reasonable for parents to respond by helping their children to escape. Indeed, mothers’ protective behaviors might appear warm and supportive (Chen, Hastings, Rubin, Chen, Cen, & Stewart, 1998; Rubin et al., 1997) and have been labeled as ‘sensitive’ and ‘responsive’. However, these behaviors may confirm the threatening nature of the stimulus. This dynamic may be especially likely to become firmly established for temperamentally fearful children, who become more physiologically aroused and emotionally distressed at lower thresholds of threat (Kagan et al., 1994). Moreover, protective, even ‘responsive’, behaviors that might otherwise be considered beneficial for children’s development could result in negative consequences for fearful children who require more practice in coping with distress independently to achieve adaptive outcomes (Arcus, 2001; Park, Belsky, Putman, & Crnic, 1997). Consistent protection may reinforce and increase their children’s fearfulness to a maladaptive level.

Intrusive Behaviors

Intrusiveness has also been linked to both early fearfulness and later anxiety. In contrast to protective behaviors that remove stress from children, intrusiveness may be characterized as pushing children toward a feared stimulus with too much force in situations involving novelty or being overinvolved in children’s tasks in situations involving challenges. Similar to protective behaviors, intrusiveness has been thought to promote the development of anxiety because this behavior takes control of the situation away from the child, preventing the development of self-efficacy and mastery over uncertain or unfamiliar situations (Chorpita & Barlow, 1998). Intrusive parenting behavior has been related to short-term increased stress for inhibited and insecurely attached children (Nachmias et al., 1996). Inhibited children of intrusive parents tend to remain inhibited or show increased inhibition across childhood (Rubin et al., 2002). Chorpita, Brown, and Barlow (1998) found that having a family environment that restricted personal control related to lower perceived control (i.e., mastery and control over events) in children, which in turn related to anxiety in a mediational framework. Messer and Beidel (1994) found that the most anxious children in their study reported the most parental control. Mothers’ intrusive involvement observed in the laboratory has been linked to anxiety in middle to late childhood (Barrett, Rapee, Dadds, & Ryan, 1996; Hudson & Rapee, 2001). Similar to protectiveness, intrusive behaviors have been conceptualized as parental reactions to children’s fearfulness or anxiety (Hudson & Rapee; Rapee, 2001).

Thus, maternal protective and intrusive behaviors have been linked to children’s fear and implicated in bidirectional interactions with anxious children, although not all mothers of fearful children react with these behaviors. It is necessary to distinguish other maternal characteristics that moderate the association between fearfulness and protective/intrusive behaviors to determine which fearful children are most at risk for this cycle. The extent to which mothers know how their children will react to uncertain situations likely affects their behavior, and being accurately attuned to their children’s fearful behavior may specifically prompt them to behave more protectively or intrusively. Although this has yet to be examined directly, the literature provides some evidence that accurate perceptions of fear strengthen the association between children’s fearfulness and maternal behavior.

Maternal Accuracy

Mothers’ perceptions of their children’s fearfulness have been linked to observed fearful temperament and may have particular influence on their parenting (Pauli-Pott, Mertsacker Bade, Haverkock, & Beckmann, 2003; Rubin et al. 1997, 1999). However, few studies to date have assessed the accuracy of these perceptions. When the match between perceptions and child behavior is assessed, it is often operationalized as the correlation between a maternal report measure and behavior observed in the laboratory. Individual differences in this type of concordance have been found specifically for children’s distress in novel situations (Leerkes & Crockenberg, 2003). Maternal report questionnaires inquire about generalities in children’s behavior that have been observed in previous situations. Therefore, one limitation of this type of concordance is that mothers and children provide information about behavior in different contexts. Moreover, how mothers report about past behavior does not necessarily reveal how they anticipate future behavior. Especially for temperamentally fearful children, a mother’s ability to anticipate fearful/shy responses in new situations seems likely to affect her behavior, which in turn shapes children’s responses in future novel situations. It remains necessary, therefore, to examine how well mothers anticipate their children’s responses in new situations.

If mothers anticipate their children’s anxious responses correctly, displays of fear may more consistently elicit protective or intrusive parenting. Several studies have examined how accurately mothers predict their children’s behavior (Casey & Fuller, 1994; Cobham & Rapee, 1999), but they did not investigate the implications of individual differences in accuracy. Closer to this end, Hastings and Rubin (1999) found that mothers who perceived their toddlers to be dispositionally fearful expected their children to be shy in hypothetical situations and endorsed comforting and protective behaviors that would probably maintain the shy behavior. These perceptions played a stronger role in predicting expected parenting behavior than did children’s observed behavior. These results suggested that when children displayed fearful temperaments and mothers perceived them accurately, mothers displayed protective behavior, but this was not tested directly. Kiel and Buss (2006) provided an initial study on individual differences in maternal accuracy in anticipating children’s fearful, sad, and angry responses. Although accuracy was not specific to fear-eliciting situations and its association to parenting behavior was not assessed, the study provided initial support for the relevance of this characteristic.

Putatively, the effect of accuracy on the association between children’s fearful temperament and maternal behavior has implications for anxiety development. Although parenting behavior has also been conceptualized as a moderator of relations between temperament and adjustment (Bates & McFadyen-Ketchum, 2000; Rubin et al., 2002), mediational models more explicitly account for evidence that child temperament can elicit parenting behavior. An increasing number of studies have shown that children affect their caregiving environments (e.g., Dumas, LaFreniere, & Serketich, 1995; Rubin et al., 1999), which is consistent with a developmental psychopathology perspective (Sroufe & Rutter, 1984). The literature therefore suggests that protective and intrusive behaviors serve as a mechanism by which early fearful temperament develops into later anxiety. Investigating how maternal accuracy affects this mechanism will contribute to our understanding of anxiety development.

The Current Study

The current study aimed to respond to limitations in the literature and was conducted with three primary goals. Firstly, the study investigated whether mothers, in general, could accurately predict how their toddlers would react in novel situations. We hypothesized that because maternal predictions were made about the same toddler behaviors that were observed by experimenters (i.e., maternal predictions and toddler behaviors occurred in the same context), maternal predictions and toddler behaviors would be significantly associated. Secondly, we aimed to understand whether accuracy moderated the association between fearful temperament and each type of parent behavior. We expected that mothers who most accurately predicted their children’s wariness would make the most attempts to change it or to help children cope with it (i.e., be protective or intrusive). Protectiveness encompassed behaviors that soothed the child or prevented the child from interacting with the stimulus. Because we focused on novelty tasks rather than challenging tasks, intrusiveness comprised behaviors that pushed the child toward the stimulus. Finally, as a third, more subsidiary goal of the study, we aimed to uncover preliminary evidence for the relevance of maternal accuracy in anxiety development. We hypothesized that moderation found in the previous aim would affect how parenting behavior mediates the longitudinal relation between fearful temperament and later anxiety. A conceptual model of this process is shown in Figure 1.

Figure 1.

A conceptual model of maternal accuracy and parenting behaviors in a developmental mechanism from early fearful temperament to later anxiety.

Method

Participants

Ninety-three mothers and their developmentally healthy two-year-old children (M = 24.76 months, SD = .42 months; 42 female) were recruited from published local birth records and participated in a larger study of toddler temperament. We sent letters to mothers about the study along with a postcard on which they could indicate their interest in participating. When their postcard returned to the laboratory, we contacted them by telephone to schedule a laboratory visit. The majority of participants were Caucasian (90 percent Caucasian, 3 percent African-American, 2 percent Hispanic, 3 percent Asian-American, and 1 percent South American Indian) and middle class (mean Hollingshead = 47.68, range = 13–66), although the sample spanned the range of socioeconomic status. A subset of the original participants (n = 61; 66 percent) completed a follow-up questionnaire battery approximately one year after their visits.

Procedure

Age 2 Laboratory Visit

Upon a mother and child’s arrival at the laboratory, the primary experimenter explained to the mother that her child would be asked to engage in a variety of activities, always in her presence. These activities were modeled after modified Laboratory Temperament Assessment Battery episodes for toddlers (Lab-TAB; Buss & Goldsmith, 2000; Goldsmith & Rothbart, 1999) and other standard laboratory procedures (Nachmias et al., 1996). Mothers received an explanation of the episodes (including what the child and mother would be asked to do and what the experimenter would do in each) and were also told they would be asked to minimize interaction with their children unless they perceived their children needed relief of distress. We used a pre-episode interview adapted from Goldsmith and Rothbart (1991), in which the experimenter asked the mother to answer several specific questions about how she anticipated her toddler would react to each episode (e.g., ‘Will your child smile at or greet the stranger?’ in the Stranger Approach episode). A complete list of questions (i.e., ‘maternal predictions’) can be found in Table 1. Mothers were instructed to answer questions, ‘Definitely yes’, ‘Probably yes’, ‘Probably no’, or ‘Definitely no’. Toddlers then participated in these episodes, which were videotaped for later scoring.

Table 1.

Maternal Predictions and Corresponding Toddler Behaviors Within Each Episode

Maternal prediction (‘Will your child …’)	Toddler behavior
Stranger approach
Smile at/greet S?	Intensity of greeting (.70)
Cry?	Intensity of distress vocalizations (.98)
Want you to hold him/her?	Intensity of attempt to be held (.89)
Be distressed when S approaches?	Intensity of distress after S’s approach (.81)
Be distressed when S talks to him/her?	Mean distress intensity after each question (.81)
Stranger working
Approach S?	Intensity of approach across episode (.95)
Want to stay close to you?	Intensity of approach towards parent (.73)
Want to play with S?	Intensity of interaction with S (.90)
Be wary of S?	Intensity of shyness across episode (.92)
Cry?	Intensity of distress vocalizations (.98)
Clown
Approach the clown?	Intensity of approach across episode (.95)
Want you to hold him/her?	Intensity of attempt to be held (.89)
Play all the games with the clown?	Number of games played (out of 3) (.93)
Be wary of the clown?	Intensity of shyness across episode (.92)
Talk to the clown?	Number of vocalizations in episode (.78)
Cry?	Intensity of distress vocalizations (.98)
Puppet show
Approach the puppets?	Intensity of approach across episode (.95).
Maternal prediction (‘Will your child …’)	Toddler Behavior
Want you to hold him/her?	Intensity of attempt to be held (.89)
Play all the games with the puppets?	Number of games played (out of 3) (.87)
Talk to the puppets?	Number of vocalizations (.78)
Be wary of the puppets?	Intensity of shyness across episode (.92)
Help the puppets when they ask?	Intensity of help behavior (.87)
Cry?	Intensity of distress vocalizations (.98)
Robot
Approach the robot?	Intensity of approach across episode (.95)
Want you to hold him/her?	Intensity of attempt to be held (.89)
Play with the robot?	Intensity of play with the robot (.80)
Be wary of the robot?	Intensity of shyness across episode (.92)
Cry?	Intensity of distress vocalizations (.98)
Spider
Approach the spider?	Intensity of approach across episode (.95)
Want you to hold him/her?	Intensity of attempt to be held (.89)
Play with the spider?	Intensity of play with the spider (.70)
Be wary of the spider?	Intensity of shyness across episode (.92)
Cry?	Intensity of distress vocalizations (.98)

Note: S = Stranger. Inter-rater reliability for the scoring of each of the toddler behaviors is located in parentheses as percent-agreement between a coder and the master coder.

Toddlers first participated in the Risk Room episode, a traditional paradigm for measuring fearful temperament (Buss & Goldsmith, 2000; Kagan et al., 1984). In the Risk Room episode, the experimenter led the child into a room with a variety of activities (i.e., tunnel, low balance beam, trampoline, box with face and cutout mouth, gorilla mask on a pedestal) and invited the toddler to play in the room for three minutes. The experimenter then returned and asked the child to try each of the activities (crawl through the tunnel, walk across the balance beam, jump on the trampoline, put a hand inside the mouth of the box, and touch the gorilla mask), giving up to three prompts to complete each activity. After completion of all activities, the experimenter ended the episode.

The next six episodes were scored to assess the match between maternal predictions and toddler fear behaviors (i.e., accuracy). In the Stranger Approach episode, the child spent 30 seconds in a room with a few small toys. A male experimenter then entered the room and attempted to engage the child in a one-minute conversation by asking short questions (e.g., ‘Are you having fun today?’), leaving afterwards. In the Stranger Working episode, the child spent 30 seconds in a room with a few small toys, after which an unfamiliar second female experimenter entered the room and sat in a desk in the corner, appearing as though she were completing work. She did not engage the child but responded to the child’s vocalizations or activity if they were directed at her. After two minutes, she left the room. In the Clown episode, the second female experimenter dressed in a clown costume (clown outfit, red nose, and colorful wig) entered the room and attempted to engage the child in three activities (blowing bubbles, catch with beach balls, and playing with musical instruments), each lasting approximately one minute. In the Puppet Show episode, the child watched two puppets (with the second female experimenter behind a small stage) engage in three activities (catch, fishing, and presentation of a sticker), which the puppets asked the child to join. In the Robot episode, the child was shown into a room in which a one-foot-tall remotecontrolled robot toy stood on a one-foot-high wooden platform. Once the child was seated in the mother’s lap, the robot moved and made noises randomly for one minute, controlled by an experimenter behind the room’s one-way mirror. After it stopped, the primary experimenter returned to the room and asked the child (with up to three prompts) to touch the robot. The experimenter ended the episode after the three prompts or the child’s contact with the robot. In the Spider episode, the child was shown into a room with a large stuffed animal spider (placed on top of a remote controlled car) in the corner. After the child was seated in the mother’s lap, the spider, controlled from behind the room’s one-way mirror, approached and withdrew from the mother and child twice, with a 10-second pause in between each movement. After it stopped, the primary experimenter returned and asked the child to touch the spider with the same protocol as in the Robot episode.

Episode order was counterbalanced so that children completed episodes in one of four randomly chosen orders. In between the aforementioned episodes, toddlers participated in non-threatening, pleasurable activities that were not scored for the purposes of the current study. Episodes alternated between two experiment rooms divided by a small room with one-way mirrors which allowed for videotaping of episodes for later scoring. All of the episodes in the current study were in the same experiment room.

Age 3 Questionnaire Mailing

We contacted mothers through a letter and response postcard around the child’s third birthday to inquire about interest in completing a battery of questionnaires. Upon return of the postcard, mothers were mailed a packet including a consent form and questionnaires about their children’s adjustment. Mothers were asked to mail the completed questionnaires back to the laboratory in a pre-stamped envelope. Thirty-two of the original 93 participants did not complete an age 3 assessment. We were unable to locate 24 families, and eight families agreed to complete the packet but never returned it.

Measures

Toddler Fearful Temperament (Age 2 Visit)

Fearful temperament was computed as a composite of the following fearful behaviors observed in Risk Room: latency to touch first toy, attempt to be held by mother, approach toward parent, tentativeness of play, and compliance with experimenter. Latency to touch the first toy was scored as the number of seconds before the child intentionally touched an object in the room. Attempt to be held, approach toward parent, and tentativeness of play were coded as intensities on a 0 (none) to 3 (strong display of behavior) scale in 10-second intervals of the episode. For each behavior, an average of the codes across the episode provided a final score. Compliance with experimenter was scored as the number of objects with which the child engaged after the experimenter’s request. Trained undergraduate and graduate research assistants scored these behaviors, and reliability with a master coder was calculated on 15–20 percent of cases. For behaviors scored in the 10-second intervals, reliability was calculated interval-by-interval, which resulted in many (roughly 50 percent) instances in which both coders gave zero-values. To account for this, we calculated percent-agreement between the coder and the master coder and found adequate reliability for the behaviors (latency to touch first toy = .90 within one second, attempt to be held by mother = .94; approach toward parent = .73, and tentativeness of play = .83). Compliance with experimenter had more evenly distributed scores and was more categorical in nature, so Cohen’s kappa was computed and reliability was found to be adequate (κ = .75). Compliance with experimenter was reverse-scored for the composite. Because the behaviors for the composite were measured on different scales, each behavior was standardized and given a z-score. The mean of these scores formed the composite, with higher values indicating more extreme fearful temperament. We examined the composite for skew and kurtosis and found that it adhered to a normal distribution.

Maternal Accuracy (Age 2 Visit)

Maternal accuracy comprised the maternal predictions and corresponding observed toddler fear behaviors for the other six episodes. Answers to the questions mothers were asked about their toddlers’ anticipated reactions to each of the episodes were scored on a 0–3 scale. Positively and negatively valenced questions were both scored so that 0 indicated the mother ‘Definitely’ predicted no distress, 1 indicated ‘Probably’ no distress, 2 indicated ‘Probably’ a reaction of distress, and 3 indicated the mother ‘Definitely’ predicted distress. The internal consistency of maternal predictions was found to be adequate (Cronbach’s alpha = .88).

Toddlers were scored for behaviors that corresponded to each of the maternal predictions. A complete list of toddler behaviors that correspond to the maternal predictions, along with the reliability scoring (i.e., percent-agreement between coders and a master coder) of each, can be found in Table 1. Behaviors were either scored as intensities or counts. The intensities of shyness and approach were scored from 1 (‘No displays’) to 5 (‘Extreme display’) considering the child’s behavior throughout a particular episode. Similarly, for intensity of distress after Stranger’s approach, one score ranging from 1 to 5 was given. Intensity of help behavior toward the puppets (0 = no attempt, 1 = approaches part way, 2 = gets ball and throws from a distance, 3 = hands ball to puppets or throws from within two feet of puppets) and intensity of play with both the robot and the spider (0 = no interaction, 1 = approaches or examines from afar with no physical contact, 2 = one instance of contact for less than 2 sec., 3 = multiple instances of contact, prolonged contact, or picks it up) were also given one rating made to encompass the child’s behavior across the episode. For average intensity of distress after each of the Stranger’s questions, one rating of distress, on a scale ranging from 1 to 5, was made after each of the stranger’s questions, and scores were then averaged to get the final score. All other intensities were given a score of 0 (no behavior shown), 1 (mild or fleeting display of behavior), 2 (definite, moderate display of behavior), or 3 (prolonged or extreme display of behavior) in each 10-second interval of the episode being scored. Undergraduate and graduate research assistants coded the behaviors. A master coder double-scored 15–20 percent of cases throughout the coding process. Percent-agreement reliability was calculated interval-by-interval for behaviors scored in 10-second intervals. For each behavior, the average of scores across these epochs yielded an overall score. We examined these scores for skew and kurtosis prior to analyses and found that they adhered to a normal distribution. All scores were standardized and given z-scores.

Maternal predictions and corresponding toddler fear behaviors from all episodes except Risk Room (scored only for toddler temperament to avoid construct overlap) were used in the construction of maternal accuracy, which is discussed in greater depth in the Results section.

Maternal Parenting Behaviors (Age 2 Visit)

Mothers were scored for various parenting behaviors (i.e., comforting, protective, and intrusive behavior) in each of the aforementioned episodes except for Risk Room, again to avoid construct overlap. Trained coders rated the maximum intensity with which mothers displayed each behavior on a 0 (none) to 3 (extreme or prolonged display) scale in each 10-second interval across each episode. Comforting behavior represented mothers’ attempts to show affection or soothe children. Protective behavior occurred when mothers shielded children from the stimulus or activity. These behaviors were scored in response to children’s bids for comfort or help. Intrusive behavior was defined as moving the child toward the stimulus in a manner inconsistent with the child’s pre-existing behavior. Each behavior was scored independently for each 10-second interval, so more than one behavior could be displayed in a given interval with varying intensities. Percent-agreement reliability (assessed interval-by-interval) for each behavior between the coders and a master coder averaged .93 for comforting behavior, .96 for protective behavior, and .95 for intrusive behavior. Consistent with previous studies of maternal protective behavior, comforting and protective behaviors were averaged within each episode and then across episodes to form protectiveness. Protective behaviors from Clown and Stranger Working were not included in this composite because they occurred with near-zero frequency. Across all other episodes, comforting and protective behavior had a strong relation (r = .84, p < .001). Internal consistency among comforting and protective behaviors across episodes was moderately strong (α = .77), suggesting that mothers enacted these behaviors similarly across episodes. The average of intrusive behavior across all episodes formed intrusiveness. The internal consistency of intrusive behaviors was low (α = .32), suggesting that mothers’ intrusive behavior differed across episodes. Despite this, we retained all intrusive behavior variables in the composite so that mothers who were more consistently intrusive had higher scores than mothers who displayed a higher level of intrusiveness in only one or two episodes.

Age 3 Child Anxiety

As part of the age 3 battery of questionnaires, mothers completed the Infant-Toddler Social and Emotional Assessment—Revised (Briggs-Gowan & Carter, 1998), which is a 193-item questionnaire that asked mothers to rate normal feelings and behaviors as well as adjustment problems they have observed in their 12–36-month-old children in the last month. Mothers respond to statements on a 0–2 Likert-style scale (0 = not true or rarely true; 1 = somewhat or sometimes true; 2 = very true or often true). The current study utilized the anxiety scales: General Anxiety (e.g., ‘Seems nervous, tense, or fearful’; 10 items), Separation Distress (e.g., ‘Cries or hangs onto you when you try to leave’; six items), and Inhibition to Novelty (e.g., ‘Is quiet or less active in new situations’; five items). The internal consistencies (Cronbach’s alpha) of the anxiety scales have all been established as above .70 for each of the scales (General Anxiety = .71, Separation Distress = .73, Inhibition to Novelty = .77) by the authors (Briggs-Gowan & Carter), and were found to be .72, .67, and .80, respectively, in the current study. The current study used a composite comprised of the mean of the three anxiety scales (α = .84 for all 21 anxiety items), hereafter referred to as age 3 anxiety.

Results

Preliminary Analyses

Order Effects

Order of episodes was counterbalanced across children to prevent consistent influence of previous episodes on later episodes. A multivariate analysis of variance was conducted to test for order effects on each of the toddler behaviors scored. Out of the 33 behaviors, only intensity of interaction with the stranger in the Stranger Working episode showed a significant effect (F = 2.97, p < .05). The small proportion of behaviors showing an order effect (3 percent) suggests that this difference may be due to chance, so order was not included in further analyses.

Attrition

The pattern of missing data for the age 3 assessment was analyzed with the Missing Value Analysis procedure in SPSS 15, and Little’s MCAR test suggested that the data were not missing completely at random (χ²[4] = 615.43, p < .05). Means of the measures from the two-year-old visit (i.e., Hollingshead index, toddler fearful temperament, maternal accuracy, and maternal parenting behaviors) were compared for the families who completed the three-year-old assessment and those who did not. Children of families who did not complete the follow-up assessment (n = 32) were somewhat higher in fearful temperament (t = 2.20, p < .05), consistent with general trends in the literature for individuals with missing data to have higher levels of maladaptive characteristics (McCartney, Burchinal, & Bub, 2006). As directed elsewhere in the literature (McCartney et al.), all longitudinal models with age 3 anxiety include fearful temperament to account for its relation with having missing data.

Current guidelines in the literature suggest that basing longitudinal analyses on only the subset of the sample who completes the later assessment (i.e., using listwise deletion) may bias parameter estimates and unnecessarily limit power, and that when the amount of attrition is at least moderate, imputation methods be used (Howell, 2007; McCartney et al., 2006). To this end, expectation/maximization treatment of missing data (i.e., the EM algorithm) was used to impute the missing values. Fearful temperament, maternal accuracy, protectiveness, intrusiveness, and existing values of age 3 anxiety were used in the algorithm. Descriptive statistics and bivariate relations among variables were calculated after imputation of missing data (Table 2).

Table 2.

Descriptive Statistics and Bivariate Relations

Variable	Mean	SD	1	2	3	4	5
1. Maternal accuracy	.08	.02	—	.55*	.22**	.35***	.31***
2. Protectiveness	.16	.21		—	.21**	.30***	.51*
3. Intrusiveness	.10	.10			—	.14	.12
4. Fearful temperament	-.02	.79				—	.56*
5. Age 3 anxiety	.59	.27					—

Note: Descriptive statistics and bivariate relations are reported for the age 3 anxiety variable after missing data were imputed with the EM algorithm.

^*p < .001

^**p < .05

^***p < .01.

Gender and SES

Comparisons of means for boys and girls were performed for fearful temperament, maternal accuracy, maternal parenting behaviors, and age 3 anxiety. T-tests revealed no significant differences in these measures between girls and boys (all p’s > .10). Bivariate correlations between the Hollingshead Index and these variables revealed no significant relations (all p’s > .10). Because of this lack of substantial differences and because investigating gender and SES were not primary aims of this study, these variables were not included in further analyses.

Maternal Accuracy

Maternal accuracy was conceptualized as the statistical strength with which mothers’ predictions actually predicted their toddlers’ corresponding fear behaviors. The nesting of maternal predictions and toddler behaviors within episodes, which were themselves nested within mother-toddler dyads, warranted examining accuracy in a three-level multilevel model (MLM; Raudenbush & Bryk, 2002). MLM analyses were conducted in several steps to investigate the properties of maternal accuracy.

Firstly, we examined a model with no predictors and toddler distress behaviors as the dependent variable. An intra-class correlation coefficient (ICC; Snijders & Bosker, 1999) was computed for toddler distress behavior at both higher levels of the model (i.e., the episode and dyad levels) to measure the proportion of total variability accounted for by within-episode and within-dyad similarity, respectively, in observations. The episode-level ICC conveyed that 28.7 percent of the variance in toddlers’ distress behaviors existed among episodes within dyads, and the dyad-level ICC showed that 15.94 percent of the variance in toddlers’ distress behaviors existed among dyads. In other words, toddler distress behaviors tended to be more similar within the same episode than across different episodes and within the same child than across children. The non-zero values at these higher levels indicated that observations of toddler distress behaviors were not independent, warranting the use of MLM.

Secondly, we entered the maternal prediction variable into the model and examined whether it required random components in addition to fixed components at the episode and dyad levels. Random components would suggest that the relation between maternal predictions and toddler distress behaviors (i.e., accuracy) may change among episodes or dyads. In other words, random components would be required to argue for individual differences in accuracy. The absence of random components would suggest the same relation between maternal predictions and toddler distress behaviors for all episodes and all dyads. A significant deviance change test (χ² [4] = 16.2, p < .01) suggested that the model fit better with these random components in addition to the fixed component, rather than only having fixed components.

The first goal of the study was achieved by examining the resulting MLM:

$${\left(\text{toddler behavior}\right)}_{\mathrm{i}\mathrm{j}\mathrm{k}}={\gamma }_{000}+{\gamma }_{100}{\left(\text{maternal prediction}\right)}_{\mathrm{i}\mathrm{j}\mathrm{k}}+\text{error}$$

The random components were located within the error term. The coefficient attached to the maternal prediction variable (γ₁₀₀) indicated the overall strength with which maternal predictions actually predicted toddler behaviors across the sample. Within this MLM, mothers’ predictions significantly related to corresponding toddler behaviors (γ = .08, t = 4.22, p < .001). Thus, mothers, in general, were reasonably successfully much of the time in predicting their toddlers’ behaviors in impending novel contexts.

Before continuing with the proposed analyses, we wanted to address one potential alternative explanation of these results. One could argue that mothers were most accurate when toddlers were simply the most fearful, and therefore, the easiest for whom to predict behavior. To investigate this more closely, the mean of all predictions and the mean of all distress behaviors were computed for each dyad, and a cross-tabulation table of a tertile split on these variables was examined (Table 3). Most ‘matches’ in this table were at medium distress, and some matches occurred at both low and high distress. Thus, it appears that it was not merely the extremely fearful children driving maternal accuracy.

Table 3.

Cross-tabulation of Low, Medium, and High Prediction and Behavior Means

	Toddler behavior mean
Maternal prediction mean	Low	Medium	High
Low	3	12	1
Medium	7	50	4
High	0	9	7

Note: Low, medium, and high groupings were based on dividing prediction and behavior means at ±1 SD around the mean.

To address the next goals of the study, we needed a score of each individual mother’s accuracy. Given the nested nature of the data, each mother-toddler dyad could be thought of as having its own regression equation with 33 (5 each in Stranger Approach, Stranger Working, Robot, and Spider; 6 in Clown; 7 in Puppet Show) observations each of maternal predictions and toddler behaviors. Just as the coefficient associated with the maternal prediction variable was interpreted as an indication of accuracy for the entire sample, this coefficient in each dyad’s regression could serve as a ‘score’ of accuracy. This score indicates the statistical strength with which a particular mother’s predictions actually predicted her own toddler’s distress behaviors. These individual coefficients were extracted as Empirical Bayesian estimates. Using this approach is consistent with our previous investigation of maternal accuracy (Kiel & Buss, 2006).

Before proceeding with core analyses, bivariate correlations among accuracy, fearful temperament, and parenting behaviors were examined (Table 2). Maternal accuracy for fear-eliciting episodes demonstrated a positive relation to toddler fearful temperament (r = .35, p < .01). In other words, mothers who correctly predicted more distress had toddlers who were also more temperamentally fearful in another setting. Maternal accuracy related to protectiveness (r = .55, p < .001) and less strongly to intrusiveness (r = .22, p < .05). Mothers demonstrating higher accuracy in anticipating their children’s responses to novelty, therefore, also tended to behave more protectively as well as more intrusively.

Maternal Accuracy in Moderation Analyses

The second goal of the current study was to examine whether maternal accuracy moderated the relation between children’s fearful temperament and parent behavior. We hypothesized that the relation between fearful temperament and parenting behaviors would be stronger as mothers displayed higher accuracy in anticipating their children’s fearfulness. In hierarchical regression analyses, maternal accuracy, fearful temperament, and their cross-product were investigated as predictors of each type of parenting behavior. Despite choosing fearful temperament as the predictor and parenting behaviors as the outcomes of these analyses, we acknowledge the likely bidirectional dynamic between the two constructs and do not intend to imply a unidirectional causality between them. Maternal accuracy and fearful temperament were centered at their means to decrease multicolinearity with and aid in interpreting the interaction term. Significant interactions were probed for simple effects by recentering predictors (Aiken & West, 1991). We centered and recentered variables by subtracting or adding the value at which the intercept was desired.

Protectiveness

A model in which the main effects of fearful temperament and maternal accuracy predicted protectiveness revealed that maternal accuracy continued to relate to protectiveness as a main effect (b = 5.08, t = 5.36, p < .001), above and beyond fearful temperament (Table 4). The addition of the fearful temperament*maternal accuracy interaction term resulted in a significant change in the model (ΔR² = .05, p < .05). The interaction significantly related to protectiveness in a positive direction (b = 2.04, t = 2.49, p < .05), consistent with expectations (see Figure 1). The interaction was probed for the relation between fearful temperament and protectiveness at various levels of accuracy (-1 SD [low], mean, and +1 SD [high]). Fearful temperament related to protectiveness significantly and positively at high maternal accuracy (b = .06, t = 2.24, p < .05), but not at mean (b = .02, t = .69, p = .49) or low (b = -.02, t = -.74, p = .46) maternal accuracy. These simple slopes are portrayed in Figure 2. The region of significance for the relation between fearful temperament and protectiveness as a function of maternal accuracy was determined using an online calculation tool (Preacher, Curran, & Bauer, 2006). Using parameter estimates and the variance/covariance matrix of the regression, the calculation tool estimates the boundaries of the region of the values of the moderator at which the predictor significantly relates to the dependent variable. The upperbound of the region for this analysis was found to be at accuracy = .09, or approximately .77 SD above the mean. Both the simple slopes calculated by recentering and the region of significance were consistent with our prediction that the dynamic between fearful temperament and mothers’ protectiveness depended on mothers having a high level of accuracy.

Table 4.

Hierarchical Regression Analyses

	Protectiveness				Intrusiveness
Variable	R2	ΔR2	b	SE	R2	ΔR2	b	SE
Step 1	.31*	.31*			.05**	.05**
Fearful temperament			.03	.02			.01	.02
Maternal accuracy			5.08*	.95			.99**	.57
Step 2	.36*	.05***			.05	.00
Fearful temperament			.02	.02			.01	.02
Maternal accuracy			4.67*	.94			.98**	.58
Fearful temp***			2.04***	.82			.04	.94
Maternal accuracy

Note: Results are provided for two models with dependent variables protectiveness and intrusiveness. N = 93 for both sets of analyses.

^*p < .001

^**p < .10

^***p < .05.

Figure 2.

Simple slopes of the relation between fearful temperament and protectiveness at Low (-1 SD), Mean, and High (+1 SD) maternal accuracy (n = 93). *p < .05.

Intrusiveness

In a main effects model, maternal accuracy only showed a trend toward relating to maternal intrusiveness (b = .99, t = 1.75, p < .10), above and beyond fearful temperament (Table 4). The fearful temperament*maternal accuracy interaction did not result in a significant change in the overall model (ΔR² = .00, p = .94), nor did it significantly relate to maternal intrusiveness (b = .04, t = .08, p = .94). Thus, mothers who more accurately anticipated their children’s fearful responses to novelty also tended to enact more intrusiveness during these tasks, but maternal accuracy did not moderate how fearful temperament related to intrusiveness.

The Role of Maternal Accuracy in Longitudinal Relations

Given a significant moderating effect in the relation between fearful temperament and protectiveness, and theoretical belief that this is important due to the relation of these variables to later anxiety, the final set of analyses explored whether and how maternal accuracy and protectiveness played a role in the prediction from toddlers’ age 2 fearful temperament to anxiety at three years of age. We hypothesized that parenting behaviors would mediate this relation and that maternal accuracy would moderate the relation between fearful temperament and parenting behavior. Following guidelines for moderated mediation analyses (Baron & Kenny, 1986; Holmbeck, 1997; Muller, Judd, & Yzerbyt, 2005; Preacher, Rucker, & Hayes, 2007), we conceptualized the following model: Path A represented the relation between fearful temperament and protectiveness (moderated by maternal accuracy), Path B represented the relation between protectiveness and age 3 anxiety, controlling for fearful temperament, and Path C represented the relation between fearful temperament and age 3 anxiety (Figure 1).

We conducted a moderated mediation analysis to investigate whether maternal accuracy moderated these three paths. Previously recommended practices for interpreting moderated mediation (Muller et al., 2005; Preacher et al., 2007) have suggested that moderation occurs for either Path A or Path B, or both, but not Path C, and that if moderation did not occur for either Path A or Path B, a main effect would still occur. Accordingly, we hypothesized that maternal accuracy would not moderate the direct relation between fearful temperament and age 3 anxiety, but rather that it would moderate either the relation between fearful temperament and protectiveness, the relation between protectiveness and age 3 anxiety, or both. Furthermore, if maternal accuracy did not moderate either of these two paths, it would be expected that a main effect would still exist, suggesting we should still find a main effect for protectiveness as a predictor of age 3 anxiety, above and beyond fearful temperament, accuracy, and their interactive effect. Finally, strong evidence for a moderated mediation would occur if fearful temperament predicted age 3 anxiety less strongly with protectiveness and all relevant moderations in the model (i.e., a significant conditional indirect effect; Preacher et al..

In the initial regression for testing this model (Path C), age 2 fearful temperament predicted age 3 anxiety (b = .19, t = 6.43, p < .001). As expected, this path was not moderated by maternal accuracy (b = 1.14, t = 1.03, p = .31). The moderation analysis conducted previously showed that fearful temperament related to protectiveness only when mothers also demonstrated high accuracy. In other words, maternal accuracy moderated Path A. To investigate maternal accuracy as a moderator of Path B, a regression model was run predicting the outcome of age 3 anxiety from protectiveness and its interaction with maternal accuracy, above and beyond fearful temperament, maternal accuracy, and the interaction between fearful temperament and maternal accuracy. The maternal accuracy*protectiveness term was not significant (b = 6.50, t = 1.47, p = .15), so it was dropped and the model was rerun. A main effect for protectiveness emerged, such that increased protectiveness predicted higher age 3 anxiety, above and beyond fearful temperament, accuracy, and their interaction (b = .54, t = 4.12, p < .001). As shown in Figure 3, with all variables included in the final model (fearful temperament, maternal accuracy, fearful temperament*accuracy, protectiveness), and accuracy centered at its mean, the relation between age 2 fearful temperament and age 3 anxiety decreased (b = .16, t = 5.21, p < .001).

Figure 3.

Moderated mediation of the relation between age 2 fearful temperament and age 3 anxiety (n = 93) with maternal accuracy centered at its mean. *p < .05, ***p < .01.

Although the ‘C’ path for this model remained significant with the inclusion of the other variables, the relation decreased in strength by 16 percent with accuracy centered at its mean. In addition, the addition of maternal accuracy, the fearful temperament*accuracy interaction, and protectiveness as predictors resulted in significant change in the overall model (ΔR² = .14, p < .001), as compared with a model in which only fearful temperament predicted age 3 anxiety.

To determine the significance of the conditional indirect effect (i.e., the value of accuracy at which the amount of the relation between fearful temperament and age 3 anxiety became significantly accounted for by protectiveness), we used the MODMED macro for estimating moderated mediation effects (Preacher et al., 2007) in SPSS 15. The results of this procedure suggested that the relation between fearful temperament and age 3 anxiety through protectiveness was significant when maternal accuracy equaled at least .10 (approximately 1.06 SD above the mean). In other words, the mediation depended on high maternal accuracy.

Discussion

Previous literature has suggested that mothers’ perceptions of fearfulness may affect the association between parenting behavior and children’s fearful temperament, but limitations have remained in the understanding of whether mothers’ accuracy in anticipating children’s fearful behavior moderates this relation. The current study aimed to investigate (1) mothers’ general accuracy in predicting toddlers’ reactions to a variety of novel situations, (2) the moderating role of accuracy in the association between fearful temperament and parenting behavior, and (3) preliminary evidence for the relevance of these variables in a moderated mediation model of anxiety development. Strengths of the study included providing a unique assessment of accuracy, using laboratory observation of the primary variables, and using the most recent statistical advances for estimating missing data, probing interactions, and exploring moderated mediation.

In general, mothers’ predictions had a strong association with toddlers’ observed distress behaviors, indicating that mothers could accurately anticipate toddlers’ reactions to novelty in the laboratory. The current study used a unique methodology for assessing the match between mothers’ expectations for their children’s responses to novelty and children’s actual behaviors in those situations. Unlike previous studies that relied on maternal report of temperament to assess concordance between mothers’ perceptions and independent observations of their children, we provided mothers with the same context as the laboratory observations of distress to form their expectations. Another improvement inherent to this methodology was inquiring about mothers’ anticipations for upcoming situations rather than about their observations of children’s previous behavior. We believe this has particular relevance for mother-child interactions in novel situations, which clearly have implications for the development of anxiety-relevant response patterns as children encounter uncertainty in the future.

Preliminary bivariate analyses revealed that mothers who demonstrated higher accuracy in predicting their children’s distress to novelty showed more protective behavior, as well as more intrusive behavior. These results are consistent with Kagan et al.’s (1984) suggestion that mothers who were most aware of their children’s fearful temperaments made active efforts to change what they perceived as inappropriately inhibited behavior. Their suggestion seemingly stemmed from the idea that this occurs because children of the intervening mothers were the most fearful, but, certainly, mothers need to be aware of and anticipate fearful behavior accurately to make the decision to intervene. Although it could be argued that accuracy was driven by the child’s fearful temperament (i.e., the most fearful children are most obvious in their behavior and so easiest for whom to form accurate anticipation), a cross-tabulation of prediction and behavior means suggested this was not the case. We argue that accuracy is more than a reflection of children’s temperament and, as a maternal characteristic, has unique influences on parent-child dynamics (Kiel & Buss, 2006). The results of the moderation analyses provided further support for this conceptualization.

As hypothesized, accuracy moderated the association between fearful temperament and maternal protectiveness. Thus, being more accurate increased the contingency between children’s dispositional fearfulness and mothers’ protective interventions. These results imply that fearful temperament does not universally elicit comforting or protective behavior from mothers. Given the wide literature demonstrating the deleterious effects of overprotection for anxiety-prone children, identifying this characteristic that increases the likelihood of this dynamic was an important result of the study.

The current study’s examination of how protectiveness mediated the relation from fearful temperament to later anxiety corroborates the previous literature on anxiety development. Although mothers may be attempting to manage their children’s distress, this protection from mild stress or uncomfortable interactions with new people or toys does not appear to be helpful in the long term. A goodness-of-fit model may be applied to parenting fearful children (see Chess and Thomas in Goldsmith, Buss, Plomin, Rothbart, Thomas, & Chess, 1987). Specifically, although responsive, comforting behavior may be beneficial for children with non-inhibited temperaments, it does not fit well with more fearful children. Although it could not be measured in the current study, the results supported previous theories that being comforting and protective in uncertain situations takes control over the environment away from the child, which, when parents behave in this way consistently, may increase children’s stress and exacerbate fearfulness (Chorpita & Barlow, 1998). It has also been suggested that this type of behavior, which externally regulates the child’s reactivity, may undermine the development of internal regulation and coping (Dadds & Roth, 2001; Sroufe, Duggal, Weinfield, & Carlson, 2000). Because the pathway from fearful temperament to protectiveness was moderated by maternal accuracy, protectiveness as a mechanism from early fearful temperament to later anxiety may not occur for mothers lower in accuracy. These results therefore augmented knowledge of when these pathways occur.

A goodness-of-fit model may also be invoked to understand why accurate appraisals of children’s fearfulness facilitated negative outcomes. Parents who are attuned to their children’s needs have typically been regarded as more sensitive, and certainly, accuracy may be beneficial with non-fear types of distress. Being able to predict how older children and adolescents will react to conflict or discipline, for example, has been shown to help parents behave in a way that facilitates compliance (Davidov & Grusec, 2006; Hastings & Grusec, 1997), arguably steering them away from externalizing trajectories. When children experience fear, accuracy may help parents choose what they think is a sensitive response (i.e., removing a feared stimulus or comforting their children), but this then feeds into anxiety development. It could be that the lack of protective behavior by inaccurate mothers may force children to develop more independent coping skills that facilitate successful regulation. Future studies examining accuracy and parenting variables along with children’s specific regulatory behaviors in novel contexts could elucidate these processes. Accuracy in general should therefore not be considered a negative characteristic, but it does appear to play a role in parent-child transactions that predict anxiety development.

The moderating role of accuracy appeared to be specific to protectiveness. Although accuracy related positively to intrusiveness, it did not moderate the relation between fearful temperament and intrusiveness. Theoretically, the accuracy we tapped into may be indicative of a type of vigilance for threat. Overprotection is a type of parenting style that has been suggested to be particularly associated with mothers’ perceptions of threat and danger (Rapee, 1997). Given this association, attunement to children’s fearful responses in potentially threatening contexts would be more relevant to protective behavior and its relation to children’s fearfulness. It would be informative to assess the relation between accuracy and other maternal variables such as anxiety, catastrophic thinking, and other cognitive characteristics in general and specific to child-rearing.

Methodological issues could also explain why relations only occurred for protective behavior. In the current study, mothers displayed a lower mean level of intrusive behavior than protective behavior. Although speculative, this may have occurred because of the instructions given to mothers to minimize their influence on their children. Perhaps protective behavior was interpreted by mothers as less influential or was more automatic among mothers than intrusive behavior and so was less susceptible to suppression. Alternatively, mothers may have been more likely to inhibit intrusive behavior because they recognized it as self-initiated, whereas if they viewed protective behavior as a necessary response to their children’s behavior, they did not interpret it as falling under the directions of the instructions. Nevertheless, intrusive behavior has been conceptualized to be deleterious to the development of internal emotion regulation (Sroufe et al., 2000), which could maintain fearfulness across childhood. Intrusive behavior and accuracy could also work together in the development of anxiety as moderators. In that case, intrusive behavior would not necessarily be expected to be related to the interaction between fearful temperament and accuracy. Therefore, the relation between maternal accuracy and intrusive parenting should be investigated in future studies that do not put such restrictions on mothers in the hopes of observing a wider range of this type of behavior.

Several limitations of the current study warrant further examination of maternal accuracy in future research. As mentioned, mothers were asked to restrict their behavior to what they considered necessary, and more variability in behavior might be achieved with more lenient instructions. The longitudinal analyses were based on the subset of the sample that completed an age 3 assessment. Attrition analyses suggested that toddlers with missing age 3 data may have been more temperamentally fearful. Although appropriate methods for imputing missing values were used, these relations should be replicated with larger samples to ensure that they hold. In addition, although we think these results provide useful information for how parents play a role in their children’s fearfulness over time in a healthy sample, future research could attempt to replicate these findings with children who are more clinically anxious. It would be informative to find out whether the mechanisms remain consistent or appear to differ. Furthermore, the outcome measure of anxiety came from a single informant, and it would be useful to broaden this to other types and informants of anxiety. Maternal report of children’s anxiety at age of three may be a different type of outcome for children than what would be observed in the laboratory or another setting such as preschool. The role of parent protective behavior as a mechanism of children’s anxiety development, therefore, cannot be generalized to other measures of anxiety outcomes without explicitly testing them. Finally, expanding this research to bigger, more diverse samples would allow for examinations of additional moderating effects, such as ethnicity and gender.

Our results suggested that maternal accuracy moderates the relation between fearful temperament and protectiveness, and that protectiveness serves as a partial mediator of the pathway between fearful temperament and later anxiety. We hope these results will inform future research examining how maternal accuracy plays a role in the development of anxiety.