Observed Profiles of Infant Temperament: Stability, Heritability, and Associations with Parenting

Abstract

Profiles of infant temperament were derived from 990 infants at 6 and 12 months of age using observed measures from the Laboratory Temperament Assessment Battery. Mothers and fathers completed questionnaires measuring parent affect and stress. Four profiles emerged at each age (typical, low negative, withdrawn/inhibited, and positive/active or low reactive) using Latent Profile Analysis. Temperament profiles show some evidence of stability and heritability, particularly for the withdrawn/inhibited group. In addition, profiles relate to parent affect and stress in different ways for mothers and fathers. Results highlight the utility of a person-centered approach to temperamental research and are discussed in relation to developmental patterns of infant temperament.

Humans classify people, behaviors, and objects into groups. In doing so, we simplify our understanding of complex relationships and decrease cognitive load in learning, decision making, and our understanding/expectations of human behavior. Temperament, or individual differences in emotional reactivity and regulation (Goldsmith et al., 1987), reflects underlying emotional and behavioral dispositions and can be observed in early infancy. The two dominant approaches in conceptualizing and measuring temperament are trait/dimension versus type/profile. Though commonly used, dimensional approaches to the study of temperament are limited; such approaches assume that individuals experience or express behaviors on a continuum (often assumed to produce a normal distribution), and that relations between dimensions are linear (Lanza, Rhoades, Greenberg, & Cox, 2011). Instead, it is possible that temperament dimensions relate in complex patterns that change as children develop.

Dimensional versus Profile Based Approaches to Temperament

Contemporary discussions of infant temperament often focus on dimensional aspects of behavior rather than profiles though this limits our understanding of the whole person. For example, height and weight are two dimensions that describe a person physically, but they must both be used to reasonably describe a person’s size, assuming that we are not specifically interested in either height or weight. In the same way, it is difficult to understand one’s typical emotional response using only one temperament dimension.

Profile based approaches to temperament use statistical methods, such as latent class/profile or cluster analysis, to derive empirically similar types of individuals across multiple potential predictors (Kagan, 2018; Loken, 2004). Using such methods, we can identify individuals who are not only quantitatively different from each other on a variable of interest, but also exhibit qualitatively different behavior. Profile based approaches mimic more ecological assessments of behavior by assuming that one trait, (e.g., activity level) may moderate the expression of another (e.g., anger proneness). Profiles also tend to highlight groups with extreme standing on any dimension of temperament in the profile. These extreme values may hold more functional significance than variants within the middle of the distribution. The argument is not that profile-based approaches should replace dimensional approaches but rather that the profile approach provides a needed complementary perspective (Kagan, 2018; Scott et al., 2016).

Identification of Temperament Profiles

Temperament profiles are increasingly being used to understand human behavior as researchers better understand the interactional nature of multiple temperament dimensions (Goldsmith et al., 1987; von Eye & Bergman, 2003). Two of the best known characterizations of temperament types come from research conducted as early as 1970. Starting at infant age two months, Thomas and Chess (1977) identified three types of infants based on parents’ reports on nine underlying temperament dimensions. The resulting difficult, easy, and slow-to-warm-up characterizations, though qualitative, are still used today. A seminal paper relating early childhood temperament to adult personality identified five temperament types (overcontrolled, inhibited, confident, reserved, and well-adjusted) from observer ratings of three year old children (Caspi & Silva, 1995). These two papers provide the basis for much of the later temperament profile research.

The following literature review focused on articles that have studied temperament using profiles. These articles, published since 1995, identify types of infant and child temperament patterns that include multiple aspects of temperament or personality (i.e., fear, anger, and positivity) in young children. The relevant literature is summarized in Table 1 with details discussed below.

Table 1.

Summary of Previous Literature Identifying Temperament Types, Classes, Profiles

Reference	Age	Measure	Number of Profiles	Profile Labels
Beekman et al., 2015	9 months 18, 27 months	Infant Behavior Questionnaire Toddler Behavior Assessment Questionnaire	4	Positive Reactive, Negative Reactive, Active Reactive, High Fear
Caspi and Silva, 1995	3, 5, 7, 9 years	Experimenter Observed Behavior	5	Under-controlled, Inhibited, Confident, Reserved, Well-Adjusted
Gartstein et al., 2017	3-8 months 9-12 months	Infant Behavior Questionnaire-Revised	3 5	Fearless/Low Positive, Frustrated/Difficult to Calm, High Positive/Regulated Low Positive, Average, Low Approach/Difficult to Calm, High Active, High Positive/Regulated
Hart et al., 1997	7 years	California Child Q-Set	3	Overcontrolled, Under-controlled, Resilient
Jansen and Matheson, 2008	18m, 30m, 4-5 years 8-9 years	EAS Temperament Survey for Children: Parental Ratings	5	Under Controlled, Confident, Unremarkable, Inhibited, Uneasy
Prokasky et al., 2008	4 years	Children’s Behavior Questionnaire	6	Unregulated, Regulated, High Reactive, Bold, Average, Well-Adjusted
Robins et al., 1996	12-13 years	California Child Q-Set	3	Overcontrolled, Under-controlled , Resilient
Sanson et. al., 2009	4 months – 12 years	Infant Temperament Questionnaire Toddler Temperament Scale Childhood Temperament Questionnaire	4	Reactive/Inhibited, Nonreactive/Outgoing, Poor Attention Regulation, High Attention Regulation
Usai et al., 2009;	28 months	Italian Temperament Questionnaire	3	Low Attention/High Motor, High Inhibition, Typical
Van den Akker et al., 2010	3-4 years	Toddler Behavior Assessment Questionnaire	3	Typical, Expressive, Fearful

Research on infant and childhood temperament profiles spans across developmental stages and cultural backgrounds. In older children, simple factor analysis identified three typologies (labeled overcontrolled, under-controlled, and resilient) across age and ethnicity in adolescent boys (Robins, John, Caspi, Moffitt, & Stouthamer-Loeber, 1996). Data from a large Norwegian sample spanning the first ten years indicated a five-profile solution using parent reported sociability, emotionality, activity and shyness (Janson & Mathiesen, 2008) and data from an Australian sample yielded four clusters of temperament types from 4 months to 12 years of age (Sanson et al., 2009). In a sample of 128 Icelandic seven-year-olds, researchers identified the same three temperament profiles and compared them to social and problem behavior during adolescence (Hart, Atkins, Fegley, Robins, & Tracy, 2003; Hart, Hofmann, Edelstein, & Keller, 1997). Usai and colleagues identified three profiles in a sample of Italian children and related them to language development in 28-month- olds(Usai, Garello, & Viterbori, 2009). These characterizations reflect not only patterns of emotional responsivity that may persist across culture and age, but also relations with cognitive and social factors that emerge later in child development.

Using quantitative assessments of infant temperament, Beekman and colleagues (2015) identified four profiles of infant temperament from parent reports of their child’s behavior (activity, anger, fear, orienting, smiling, soothability) at 9-, 18-, and 27-months of age. They termed these profiles as positive reactive, negative reactive, active reactive, and high fear (Beekman et al., 2015). Prokasky and colleagues (2017) used parent reported temperament across three studies to identify and validate the formation of a six cluster solution of unregulated, regulated, high reactive, bold, average, and well-adjusted children approximately four years old (Prokasky et al., 2017). Van den Akker, Deković, Prinzie, and Asscher (2010) identified three profiles of infants using mother reported sociability, anger and activity level from 30 to 40 months of age. These three profiles were labeled as typical, expressive, and fearful. Finally, Gartstein et al. (2017) used pooled data from eight different research groups to find evidence for a three-profile model in younger infants (3 to 8 months) and a five-profile model for older infants (9 to 12 months). Their profile labels reflect high or low levels of specific temperament dimensions, for example Low Positive or Frustrated/Difficult to Calm.

In sum, several recent works have identified multiple temperament profiles in early infancy and childhood. Longitudinal solutions indicate stability in temperament profile across time in the Norwegian and Australian samples (Janson & Mathiesen, 2008; Sanson et al., 2009). Each of the aforementioned studies found between three to six qualitatively different types of infants and children. Notably, the number of profiles differed depending on age, type of measurement and sample size, yet each paper identified a well-regulated temperament type and an unregulated temperament type, with slightly different labels. Many of the analyses also identified an inhibited or fearful profile.

Stability of Temperament Profiles

Temperament is moderately stable (Goldsmith et al., 1987; Rothbart & Bates, 2006; Rothbart, Ahadi, & Evans, 2000; Thomas & Chess, 1977). Meta-analytic data examining stability in infancy shows cross-time correlations for temperament dimensions to be ρ = .35 from birth to approximately 3 years of age, across dimension and measurement method (Roberts & DelVecchio, 2000). We might expect, then, for temperament profiles that characterize the whole person to also be stable across time. Indeed, profile approaches have provided some evidence for infant temperamental stability across time, with mixed findings across age and typology. Although some similarity was apparent across time in all four profiles found by Beekman et al. (2015), stability was more apparent in the more reactive profiles, such that 9-month-old individuals classified as Positive Reactive and Negative Reactive were 78% and 51% (respectively) more likely to also be Positive Reactive and Negative Reactive at 27 months of age. Infants who at 18 months were classified as Fearful were more likely to remain in the Fearful group at 27 months. Janson and Mathieson (2008) observed significant stability across four waves of measurement: 44% of children stayed in the same temperament type from 18 to 30 months; 46% of children maintained the same temperament type from 30 months to 4–5 years; and 33% of children maintained the same temperament type from 4–5 years to 8–9 years. Stability was the same for boys and girls (Janson & Mathiesen, 2008). Van den Akker et al. (2010) found that approximately 70% of infants maintained the same profile membership across four measurements. In sum, evidence suggests moderate stability in temperament profile membership, a finding similar to those for dimensional approaches to observed and parent reported temperament (Planalp, Van Hulle, Gagne, & Goldsmith, 2017; Putnam, Ellis, & Rothbart, 2001; Putnam, Rothbart, & Gartstein, 2008). Importantly, each of the studies used parent reported temperament, perhaps enhancing stability estimates by neglecting the interdependent nature of parent and child behaviors across time as well as method variance due to reporter bias.

Heritability of Infant Temperament

To our knowledge, no work has examined twin similarity in temperament profiles using a broad constellation of temperament constructs. However, extensive evidence indicates that MZ twins are more similar than DZ twins for most temperamental traits (Gagne, Vendlinski, & Goldsmith, 2009). Further, there is more within twin pair similarity in negative aspects of temperament than positivity or activity level (Goldsmith, Lemery, Buss, & Campos, 1999). Indeed, in work using the same sample as this study that only examined infant fear, MZ twins were more likely than DZ twins to be classified in the same fear trajectory profile (Brooker et al., 2013). In two other papers, moderate heritability was estimated for lab-assessed anger (Gagne & Hill Goldsmith, 2011) but there was no genetic effect for infant positivity (Planalp, Van Hulle, Lemery-Chalfant, & Goldsmith, 2017). Though not a focus of this paper, we will also examine within twin pair similarity of profile membership to determine if some profiles perhaps reflect a heritability that mimics those found with individual temperament dimensions.

Parenting and Temperament Profiles

Family systems theories of development posit that, within a family, traits of one member interact with and affect traits of another member (Cox & Paley, 2003). Extant research examines parent behaviors and infant temperament, though many of these studies focus on a single dimension: i.e., parents reporting higher stress levels have infants rated higher in stranger fear (Brooker et al., 2013); parent positivity predicts infant smiling and laughter (Bridgett, Laake, Gartstein, & Dorn, 2013; Planalp et al., 2017). However, we conceive of parents interacting with the whole rather than simply with one aspect of the infant’s emotionality, which suggests the value of using a profile approach toward infant temperament in studies of parenting effects.

Combining a family systems approach with work identifying infant temperament profiles, self-reported parent personality was related to infant temperament profiles such that, in general, more negative mothers and fathers had infants classified as Negative Reactive; more reward dependent mothers and fathers had infants classified as Active Reactive (Beekman et al., 2015). This study used both adoptive and birth parent reports in a large sample. Notably, adoptive parent reports of their own personality were related to their reports of infant temperament; the same was not found for birth parents’ reports of their own personality. This, perhaps, indicates that parents are more likely to rate their children as similar to themselves, whether genetically related or not, and highlights the potential utility of using less biased observer ratings of infant temperament. Van den Akker and colleagues (2010) also found that parenting related to child temperament profile membership, such that children classified as Typical had parents who used more positive parenting and Fearful children had parents who used less positive and more negative parenting. This finding could reflect reporter bias whereby positive or negative parents rate their children similar to themselves. To better understand unique contributions relating parenting to infant temperament, we use laboratory observed assessments of infant temperament to identify profiles unaffected by reporter bias.

Previous work on temperament profiles in infancy relies on maternal report, which may bias the measurement of infant behaviors. Parents may compare siblings’ or peers’ temperaments rather than focus solely on discrete behaviors by one child. This is apparent in twin research using parent reports; parent reports enhance differences in non-identical (dizygotic; DZ) twins more so than identical (monozygotic; MZ) twins (Goldsmith & Hewitt, 2003; Goldsmith & Rothbart, 1991; Goldsmith et al., 1999, p. 199; Saudino, 2003). To our knowledge, only Caspi and Silva (1995) used observational data to identify temperament types from 3 to 9 years of age. Other work (Hart et al., 1997; Robins et al., 1996) used a Q-sort rated by multiple judges but raw data were extrapolated from transcripts, not from behavioral observations. Using scores derived from observed behavioral paradigms which elicit temperament provide, ideally, accurate reports of infant temperament which are unbiased by parent personality or sibling comparisons (Goldsmith & Gagne, 2012).

Research Questions

We conducted latent profile analysis (LPA) to identify typologies of temperament at six and 12 months of age. LPA uses the underlying latent structure of the data to identify patterns of behavior across multiple variables for each individual. Profiles are characterized by between and within patterns of behaviors; that is, temperament dimensions are expressed within a person, but differently between the identified groups of individuals. Each individual is assigned the probability of being classified into each profile (Bauer & Curran, 2004). Rather than relying on parent reported temperament, we observed infant behavior from laboratory assessments to identify profiles of early infant temperament. Similar to previous work, we expected to find between three to six temperament profiles by statistically comparing several LPA models. In addition, we expected to find a controlled or regulated profile as well as a more inhibited profile. Further, we expected profiles reflecting more extreme traits to be more stable (Beekman et al., 2015; Biederman et al., 2001). Because we utilize a sample of twins, we also conducted genetic analyses to examine the degree to which MZ twins were similar or different in profile membership compared with DZ twins. Finally, and adding to existing literature on temperament and parenting, because infant behaviors are often associated with parenting and the environment, we examined associations of infant temperament profiles with parent characteristics (affect and stress).

Method

Participants

Data were drawn from a study of twins examining the genetic and environmental underpinnings of temperament through early childhood. Families were recruited using multiple methods, including state birth records, newspaper birth announcements, television advertising, and flyers in doctors’ offices (Lemery-Chalfant, Goldsmith, Schmidt, Arneson, & Van Hulle, 2006). Parents and their infant twins (N = 990 infants; 494 families, 2 families had useable data for only 1 twin) were assessed in both the laboratory and with questionnaires. Demographic measures indicate 92.9% of infants were Caucasian (3.2% Hispanic), 3.6% of infants were African-American, and 1.4% were Asian-American. Approximately half the infants (n = 506, 51.1%) were female. Mothers were on average 32 years old (M = 31.83, S.D. = 4.72) and fathers were 34 years old (M = 33.65, S.D. = 5.70), with median family income above $50,000. Seventy-five percent (75.1%) of mothers and 76.6% of fathers had completed college and 14.3% and 19.7% of mothers and fathers, respectively, had only a high school education. Twin zygosity was determined using parent report, observer ratings, and the genotyping of ambiguous cases. The Zygosity Questionnaire for Young Twins (Goldsmith, 1991) has a greater than 95% agreement with genotyping (Forget-Dubois et al., 2003; Price et al., 2000). In this sample, we have 331 MZ twin pairs, 369 same sex twin pairs, and 280 opposite sex twin pairs. Demographics are in Table 2.

Table 2.

Participant Demographics

Infant Sex and Zygosity by Sex		n = 506 female
	MZ=172	SSDZ=188	OSDZ=140
		n = 484 male
	MZ=159	SSDZ=181	OSDZ=140
Infant Age at first (6 month) visit	Mean Age: 26.71 weeks (S.D. 1.65, range 22-33 weeks)
Infant Age at second (12 month) visit	Mean Age: 53.06 weeks (S.D. 2.07, range 49-62 weeks)
Infant Racial Background
Caucasian/White	920 (32 Hispanic)
African American/Black	36
Asian	14
American Indian	6
Family Income	Median = $51,000 - $60,000
Mother Age at infant birth (years)	Mean: 31.82 (S.D. 4.72)
Father Age at infant birth (years)	Mean: 33.65 (S.D. 5.70)
Mother Education (years)	Mean: 15.50 (S.D. 2.38)
Father Education (years)	Mean: 15.35 (S.D. 2.59)

Note: Total N = 990. Due to missing questionnaire data, sample sizes for some demographic variables varied. MZ: monozygotic; SSDZ: same sex dizygotic, OSDZ: opposite sex dizygotic.

Procedures

Infants participated in laboratory assessments of temperament at 6 and 12 months of age. All measures (infant temperament, parent affect, and parenting stress) were completed at both ages. Visits took approximately one hour. Parents completed questionnaires separately and mailed them back to the lab. Average time between observed infant temperament in the lab and parent questionnaire completion was 2.20 weeks for mothers and 2.04 weeks for fathers at 6 months; average time difference was 2.72 weeks for mothers and 2.83 weeks for fathers at 12 months.

Measures

Laboratory Temperament Assessment Battery.

The Laboratory Temperament Assessment Battery (Lab-TAB; Goldsmith & Rothbart, 1996) is an assessment designed to measure infants’ reactions to stimuli that elicit emotional or behavioral reactions across five broad dimensions of infant temperament: fearfulness, anger proneness, joy/pleasure, interest/persistence, and activity level (Planalp, Van Hulle, Gagne, et al., 2017). We used two episodes measuring each broad temperament dimension (fear, anger, sadness, positive affect) and one episode each for interest and activity level from the Pre-locomotor and Locomotor versions of the Lab-TAB (Goldsmith & Rothbart, 1996). Lab-TAB episodes were videotaped for later scoring and coders were trained by a master coder. A complete description of procedures, scoring, and reliability estimates for the sample are available in Planalp et al. (2017), with brief descriptions below.

Fear: Stranger Approach.

Social fear is elicited when the infant sits in a high chair across the room from a door where a male experimenter unknown to the infant enters. The stranger first enters the room and waits for 10 seconds. Then the stranger slowly (10 seconds) approaches the infant, saying “Hello…I’m going to come a little closer to you now.” The stranger nears the infant, kneels and stares at the infant for 30 seconds. Behaviors are scored by trained coders for each of the three stages. Scores for each stage include facial and bodily fear (0–3 Likert scale), facial sadness (0–3 Likert scale) and bodily sadness (0/1), distress vocalizations (0–5 Likert scale), and escape behaviors (0–3 Likert scale). Average coding reliability for the Stranger Approach was ĸ = .80.

Fear: Masks.

Object fear is elicited by presenting the infant with four increasingly scary masks for 10 seconds each: an evil cartoon queen, an old man, a vampire, and a gas mask. Like Stranger Approach, several behaviors are scored during each “mask” trial: facial and bodily fear (0–3 Likert scale), facial (0–3 Likert scale) and bodily (0/1) sadness, distress vocalizations (0–5 Likert scale), and escape behaviors (0–3 Likert scale). Average coding reliability for the Masks episode was average ĸ = .76.

Anger and Sadness: Arm Restraint.

During Arm Restraint, mothers hold their infant arms down to prevent him/her from playing with an enticing, shiny toy. From two trials of 30 seconds each, we score bodily struggle (0–4 Likert scale), facial anger and facial sadness (0–3 Likert scale), distress vocalizations (0–5 Likert scale), and bodily sadness (0/1). Average coding reliability for Arm Restraint was ĸ = .79.

Anger and Sadness: Car Seat.

A second anger/sadness episode elicited distress as the infant was placed in a car seat by the parent and physically restrained for a total of 30 seconds. Like Arm Restraint, coded behaviors include facial anger (0–3 Likert scale), bodily anger (0–4 Likert scale), and distress vocalizations (0–5 Likert scale). Facial sadness (0–3 Likert scale) and bodily sadness (0/1) are also scored during the Car Seat episode. Average coding reliability for Car Seat was ĸ = .77.

Positive Affect: Puppets.

Experimenters played a puppet game with infants to elicit positivity from the infant. The experimenter talks to the infant using fun voices, tickles the infant, and then gives the puppets to the infant to play with for 2 minutes. From this interaction, we score smiling, laughter, positive vocalizations and positive motor acts 0 ‘not present and 1 ‘present.’ Smiling intensity is scored on a Likert scale ranging from 0 ‘no smiling’ to 3 ‘large smile.’ Average Puppets coding reliability was ĸ = .85.

Positive Affect: Peekaboo.

Mothers play the classic “peekaboo” game with their infants while standing behind a screen. Like the Puppets episode, several items are scored during this mother-infant play: smiling, laughter, positive vocalizations and positive motor acts (0/1) and intensity of smiling (0–3 Likert scale). Coding reliability for Peekaboo had an average ĸ = .86.

Attention: Slides.

During the Slides episode of the Lab-TAB, the infant is oriented toward a screen where he/she views a series of 15 slides for various time increments. The slides were selected to elicit interest or attention. Examples include picture of a colorful flower, a girl playing, and a mother and baby bird. Slides were presented in the same order for all infants. The first slide is presented for 2 seconds, the second slide is presented for 4 seconds, the third slide is presented for 6 seconds, the fourth slide is presented for 8 seconds and finally the last slide of the trial is presented for 10 seconds. This is repeated three times totaling 90 seconds. Scored behaviors include interest (0–2 Likert scale), duration of looking (0–3 Likert scale), presence of gestures (0/1) and vocalizations (0/1). Coding reliability estimates for Slides ranged from .79 to .88, with an average ĸ = .85.

Activity Level: Basket of Toys.

During Basket of Toys (6 months only), the infant sits (supported) beside a basket of toys designed to be engaging for 6-month-olds for 3 minutes. Coders score the intensity of the infant’s manipulation of the toys (0–4 Likert scale) and the number of different toys the infant plays with during each epoch. Average scoring reliability was ĸ = .67.

Free Play.

During the Free Play episode (12 months only), infants played in a room filled with toys, such as a truck, balls, inflatable rings and a drum. This is similar to the Basket of Toys but allows more movement around the room and toys that are developmentally appropriate for a 12-month-old. Five minutes of free play was coded for intensity of manipulation of toys (0–4 Likert scale), number of toys played with plays with, intensity of movement (0–4 Likert scale) and number changes in body position. Average coding reliability was ĸ = .72.

Parent Affect.

We measured parent affect using the Positive and Negative Affect Scale (PANAS; Watson, Clark, & Tellegen, 1988). The PANAS assesses the frequency of emotional expression using 20 questions, such as “How often in the last few weeks have you felt … distressed, excited, upset, ashamed, alert?” on a 1–5 Likert scale ranging from “Very slightly or not at all” to “Extremely.” Items are average to create a positive affect composite, which includes emotions such as interest, excitement, strength, enthusiasm, pride, alertness, activity and inspiration. The negative affect composite includes emotions such as distress, guilt, fear, hostility, irritability, and shame. Mothers and fathers completed the questionnaire separately. Average reliability for mothers for the PANAS was α = .87 for the positive scales and α = .87 for the negative scales at 6 and 12 months. For fathers, average reliability was α = .87 for the positive scales and α = .86 for the negative scales at 6 and 12 months.

Parenting Stress.

Parents completed a shortened form of the Parenting Stress Index (PSI; Abidin, 1990) to assess their own level of stress relating to parenting. We used 38 PSI questions representing parent characteristics and situational factors related to stress. Parents responded to items such as “Since I brought my child home from the hospital, I find that I am not able to take care of him/her as well as I thought I could. I need help” on a 5-point Likert scale from “Strongly Disagree” to “Strongly Agree.” Mothers and fathers completed the questionnaire separately. Reliability estimates for mothers was α = .84 at 6 months and α = .83 at 12 months. For fathers, reliability was α = .84 and .85, at 6 and 12 months respectively.

Missing Data

Recruitment was ongoing throughout the course of the study, so more infants participated in the 12-month (n = 915) than 6-month assessment (n = 594), and missing data occurred at each age (see Table 3 for sample sizes). If data were missing for one twin, they were likely also missing for the other twin within the same family; thus, we conducted two missing values analyses using SPSS Version 22 by randomly selecting first one twin then the other to test for non-random missingness at the family level. Little’s MCAR (Missing Completely at Random) Test incorporating all study and demographic variables was nonsignificant (χ² _{(df = 750)} = 667.16, p = .99 for twin 1; χ² _{(df = 738)} = 649.59, p = .99 for twin 2). Patterns of missing data were not systematically biased due to any variables of interest in our study, and further analyses used maximum likelihood estimation (MLE; Enders, 2010), a robust method for calculating estimates using all available data.

Table 3.

Descriptives and Sample Sizes for Study Variables

		Lab-TAB Fear	Lab-TAB Anger	Lab-TAB Sadness	Lab-TAB Positive Affect	Lab-TAB Interest	Lab-TAB Activity Level
6 months	N (individuals)	575	564	563	565	536	369
	Mean	0.01	0.02	0.01	0.00	−0.01	0.00
	SD	0.55	0.70	0.66	0.54	0.58	0.91
12 months	N (individuals)	899	878	878	894	877	837
	Mean	0.00	0.01	0.01	−0.01	−0.01	0.00
	SD	0.56	0.65	0.67	0.57	0.59	0.66
		Mother Positive Affect	Mother Negative Affect	Mother Parenting Stress	Father Positive Affect	Father Negative Affect	Father Parenting Stress
6 months	N (families)	228	228	230	214	214	219
	Mean	3.75	1.90	2.33	3.58	1.84	2.30
	SD	0.63	0.59	0.45	0.63	0.60	0.42
12 months	N (families)	320	320	323	292	292	294
	Mean	3.79	1.87	2.33	3.57	1.84	2.29
	SD	0.64	0.63	0.44	0.62	0.60	0.44

Note: Lab-TAB = Observed temperament from the Laboratory Temperament Assessment Battery. Ns for infant variables reflect individual infants and vary according to how many infants completed an episode; Ns for parents reflect number of parents within a family who completed questionnaire data; thus, parent Ns are smaller than infant Ns in this sample of families with twins.

Infant Gender.

Previous work from this sample (Planalp et al., 2017) indicates that girls exhibit more fear and boys exhibit increased interest and activity level, as measured dimensionally. Therefore, we included gender, not as an indicator of class membership, but a covariate in profile derivations.

Results

Basic correlations and descriptive statistics for observed temperament are in Tables 3 and 4. For Lab-TAB temperament ratings, item level scores are standardized before computing a summary or dimension score (Planalp et al., 2017). In general, temperament dimensions that reflect negativity, specifically fear, anger, and sadness, were related within and across time (see Table 4). However, patterns of significant correlations between temperament dimensions varied at 6 and 12 months; for example, activity level was related to other dimensions at 6 months, but no longer related to anger at 12 months. In addition, all temperament dimensions showed significant stability from 6 to 12 months (bottom panel of Table 4).

Table 4.

Correlations within and across Lab-TAB Observed Temperament Dimensions

6 months	1	2	3	4	5	6
1. Fear	1
2. Anger	.25***	1
3. Sadness	.23***	.44***	1
4. Positive Affect	−.11*	.01	.00	1
5. Interest	−.01	−.07	.00	.02	1
6. Activity Level	.11*	.12*	.27***	.27***	−.01	1
12 months	1	2	3	4	5	6
1. Fear	1
2. Anger	−.02	1
3. Sadness	.16***	.39***	1
4. Positive Affect	−.11**	.12**	.00	1
5. Interest	−.05	−.08*	−.07†	.11**	1
6. Activity Level	−.02	.13***	.21***	.17***	.08*	1
Correlations across time	Fear (12m)	Anger (12m)	Sadness (12m)	Positive Affect (12m)	Interest (12m)	Activity Level (12m)
1. Fear (6m)	.17***	.03	.06	.01	.07	.00
2. Anger (6m)	.05	.24**	−.13***	−.05	.00	.05
3. Sadness (6m)	.06	−.13**	.38***	.09*	−.05	−.01
4. Positive Affect (6m)	.01	.05	.12**	.18***	.01	.07
5. Interest (6m)	.03	−.07†	.00	−.01	.13*	.02
6. Activity Level (6m)	.10	.04	.18***	−.05	−.04	.32***

Note:

^†p < .10,

^*p < .05,

^**p < .01,

^***p < .001.

Profile Selection

We tested a series of latent profile models using MPlus 7.3 (Muthen & Muthen, 1998) to construct profiles of infant temperament across all six dimensions (fear, anger, sadness, positive affect, interest, and activity level). LPA identifies subgroups of individuals that differ from the overall average of a sample of individuals. Using the six continuous indicators of temperament dimension at each age, we compared models identifying one to six profiles (k).

Figure 1 provides an illustration of the model specified. We ran models with 1000 random starts, assumed a normal distribution, covariances between indicator items (σ_x∣y) in each class were fixed to zero, means (μ) and variances (σ²) for each indicator item within each class were freely estimated. Indicator items within each class were correlated with each other but this is not depicted in Figure 1 for simplicity. In addition, profile analyses included a clustering variable for family, so that analyses would appropriately account for within twin pair similarity when estimating the covariance for within twin pair observations.

Figure 1.

Illustration of latent profile model

In the LPA literature, debate continues on the best process for profile selection, with both theory and the given data driving researchers’ decisions to choose the correct number of profiles. The most robust model selection criteria are the Bayesian Information Criteria (BIC; Schwarz, 1978) and the adjusted Vuong-Lo-Mendell-Rubin Likelihood Ratio Test (VLMR; Lo, Mendell, & Rubin, 2001). The BIC and VLMR comparison statistics have the strongest power to detect an accurate number of classes given the data (Tein, Coxe, & Cham, 2013). A lower BIC is preferable to a higher one, and the VLMR compares the k-class vs. k-1 class models using a Chi-Square distribution. Fit statistics for each LPA model at 6 and 12 months of age are in Table 5. Given our data, the 4-profile mixture fit the data best at both ages. Though previous work has found between 3–6 profiles, these have all been with parent reported data and at slightly different ages. Thus, we relied on statistical criteria for selection of the number of profiles rather than trying to replicate any particular previous finding.

Table 5.

Fit indices and class membership for latent profiles of infant temperament at 6 and 12 months.

6 Month Temperament Profiles
Number of Profiles (k)	1	2	3	4	5	6
Number of free parameters	29	41	55	69	83	Did Not Converge
LogLikelihood Value	−5788.60	−4029.91	−3942.25	−3897.80	−3871.64
BIC	11777.23	8321.68	8235.77	8236.29	8273.39
Entropy	n/a	.72	.70	.69	.71
Class Membership	594 (100%)	390 (66%)	252 (42.5%)	216 (36%)	212 (36%)
		204 (34%)	215 (36%)	203 (34%)	204 (34%)
			127 (21.5%)	111 (19%)	74 (12.5%)
				64 (11%)	62 (10.5%)
					42 (7%)
Comparing models with k to k-1 classes
VLMR Adjusted LRT	n/a	700.05, p<.0001	173.38, p=.22	87.92, p<.05	51.73, p=.22
12 Month Temperament Profiles
Number of Profiles (k)	1	2	3	4	5	6
Number of free parameters	29	41	55	69	83	97
LogLikelihood Value	−8730.36	−7048.65	−6768.70	−6675.22	−6636.35	−6607.42
BIC	17660.76	14376.88	13912.44	13820.96	13838.67	13878.27
Entropy	n/a	.58	.75	.73	.70	.72
Class Membership	915 (100%)	628 (69%)	544 (59.5%)	402 (44%)	329 (36%)	330 (36%)
		287 (31%)	189 (20.5%)	187 (20.5%)	193 (21%)	193 (21%)
			182 (20%)	186 (20.5%)	176 (19%)	126 (14%)
				140 (15%)	119 (13%)	119 (13%)
					98 (11%)	96 (10%)
						51 (6%)
Comparing models with k to k-1 classes
VLMR Adjusted LRT	n/a	548.18, p<.05	554.10, p<.0001	185.011, p<.001	76.95, p=.20	55.29, p=.06

Note: BIC = Bayesian information criteria; VLMR = Vuong-Lo-Mendell-Rubin Likelihood Ratio Test; the 4 Profile model (bolded) was selected at both ages.

Temperament Profiles

MPlus provides means and variances for each of the six indicator variables for each profile. Figure 2 depicts the means from the best fitting 4-profile model at each time point. At 6 months, the typical profile (n = 203, 34% of the sample) reflects infants who are lower in activity level and distress (fear/sadness) yet show slight anger at 6 months. The low negative profile (n = 216, 36%) includes infants who are lower in anger and activity level but slightly higher in fear. The withdrawn/inhibited profile (n = 111, 19%) includes infants with slightly higher activity level but who are also high in the sadness dimension. Finally, the positive/active profile (n = 64, 11%) includes infants who are more reactive and more positive. At 12 months, these profiles were mostly replicated with similar profile descriptions. There was a typical (n = 402, 44%), a low negative profile (n = 187, 20%), and a low reactive profile (n =186, 21%). At 12 months, infants in the low reactive profile exhibited similar patterns of behavior as the 6-month positive/active profile, but with lower levels of activity level. The withdrawn/inhibited profile also recurred at 12 months, but infants in this profile also exhibited slightly higher fear (n = 140, 15%).

Figure 2.

Profiles of observed temperament at 6 (2a) and 12 (2b) months of age

MPlus assigns each individual a posterior probability of profile membership, somewhat analogous to a factor score in simple factor analysis (Muthen & Muthen, 1998). For example, an individual may be assigned to Profile 3, but their posterior probabilities of profile membership indicate likelihoods of being assigned to each of the 4 profiles, and these probabilities sum to 1.0 (i.e., Profile 1: 0.05; Profile 2: 0.10; Profile 3: 0.80; Profile 4: 0.05). In further analyses, we used posterior probabilities as continuous indicators of profile membership instead of ordinal class membership. Ordinal class membership can introduce error by assuming complete accuracy in profile assignment (Berlin, Williams, & Parra, 2014).

Stability of Profile Membership Probability

Posterior probabilities of profile membership were used to examine cross-time associations in profile membership. Models used multilevel mixed methodology that specified a compound symmetric covariance matrix to account for the nonindependence of assessing twins within families (Sayer & Klute, 2005). The probability of being classified in the same profile at both 6 and 12 months was significant for three of the four profiles (β = .08, p < .001 for typical; β = .17, p < .001 for low negative; β = .34, p < .001 for withdrawn/inhibited; β = −.02, p = .70 for positive/active with low/reactive). Of note, the probability of being in the withdrawn/inhibited profile at 6 months was inversely related to being in the low negative profile at 12 months, and the probability of being in the withdrawn/inhibited profile at 12 months was inversely related to being in the typical and low negative profiles at 6 months. The full set of estimates of associations across profiles are in Table 6.

Table 6.

Stability and Comparison of Profile Membership Probability

		12 Month Profile Membership Probability
		Profile 1 / Typical	Profile 2 / Low Negative	Profile 3 / Withdrawn/Inhibited	Profile 4 / Low Reactive
6 Month Profile Membership Probability	Profile 1 / Typical	.08**	.05	−.20***	.02
	Profile 2 / Low Negative	−.05	.17***	−.22***	.05
	Profile 3 / Withdrawn/Inhibited	−.05	−.16***	.34***	−.05
	Profile 4 / Positive/Active	.01	−.05**	.08*	−.02

Note:

^†p < .10,

^*p < .05,

^**p < .01,

^***p < .001. Estimatesm from multilevel mixed models accounting for nonindependence of twin within a family.

Twin Similarity of Temperament Profiles

We examined within twin pair similarity of temperament profile membership for monozygotic (MZ) and dizygotic (DZ) twins using intraclass correlations (ICCs) and univariate ACE models comparing the continuous, probability of profile membership variables. ICCs indicate the relative similarity of a phenotypic trait within MZ and DZ twin pairs. A greater ICC for MZ than DZ pairs indicates genetic variance for the trait. Results show a higher MZ correlation than DZ correlation for the typical, low negative and withdrawn/inhibited profiles, indicating some heritability for the expression of multiple temperament domains. There was no similarity within twin pairs for the positive/active profile at 6 months or the low reactive profile at 12 months (see Table 7).

Table 7.

Twin Similarity (Intraclass Correlations) for Temperament Profiles

Profile Label	6 Month ICCs		12 Month ICCs
	MZ	DZ	MZ	DZ
Typical	.34***	.13†	.31***	.16**
Low Negative	.37***	.24***	.16†	.11†
Withdrawn/Inhibited	.46***	.27***	.50***	.36***
Positive/Active and Low Reactive	.00	.09	−.02	.22**

Note:

^†p < .10,

^*p < .05,

^**p < .01,

^***p < .001

Using Structural Equation Modeling in MPlus, we fit univariate twin models to the posterior probabilities of profile membership. We transformed the posterior probabilities using a square root function and used maximum likelihood estimation, which more appropriately adjusts for the non-normality of skewed data. The full (ACE) and reduced (AE, CE, and E) models decompose the phenotypic variance of profile membership into additive genetic (A, h²), shared environment (C, c²) and nonshared environment (E, e²) components. Knopik et al. (2016) provide a full description of this method. The reduced models are nested within the full model, allowing us to select the best fitting model given the data for each profile. Model fit statistics and parameter estimates for genetic, shared and nonshared environmental effects are presented in Table 8. In keeping with the ICCs, the withdrawn/inhibited profile at both ages showed a strong genetic (A, h²) influence with little contribution from the shared environment (C, c²). The best fitting model for the positive/active and low reactive profiles was the reduced E model, indicating a strong nonshared environment effect and indicting little similarity across twins in this profile.

Table 8.

ACE Model Fit and Estimates of Genetic, Shared Environment, and Nonshared Environment Contributions for Temperament Profiles

6-Month Profile	MODEL	BIC	−2LL	# free	χ2	df	Δχ2	Δdf	p	h2	c2	e2
Typical	ACE	461.90	439.12	4	1.68	6	--	--	0.947	0.382	0.000	0.618
	AE	456.21		3	1.68	7	0.00	1	0.975	0.382	--	0.618
	CE	460.04		3	5.52	7	3.83	1	0.597	--	0.256	0.745
	E	474.19		2	25.36	8	23.67	2	0.001	--	--	1.000
Low Negative	ACE	664.10	641.32	4	0.29	6	--	--	1.000	0.176	0.175	0.650
	AE	659.64		3	1.53	7	1.23	1	0.981	0.393	--	0.607
	CE	659.07		3	0.96	7	0.67	1	0.996	--	0.293	0.707
	E	679.34		2	26.92	8	26.63	2	0.001	--	--	1.000
Withdrawn/Inhibited	ACE	399.26	376.48	4	9.08	6	--	--	0.169	0.307	0.154	0.539
	AE	394.65		3	10.16	7	1.08	1	0.180	0.493	--	0.507
	CE	395.92		3	11.43	7	2.35	1	0.121	--	0.359	0.640
	E	429.98		2	51.18	8	42.10	2	0.000	--	--	1.000
Positive/Active	ACE	172.78	150.00	4	16.14	6	--	--	0.013	0.000	0.053	0.947
	AE	167.65		3	16.70	7	0.56	1	0.019	0.042	--	0.958
	CE	167.09		3	16.14	7	0.00	1	0.024	--	0.053	0.947
	E	162.22		2	16.97	8	0.83	2	0.030	--	--	1.000
12-Month Profile	MODEL	BIC	−2LL	# free	χ2	df	Δχ2	Δdf	p	h2	c2	e2
Typical	ACE	766.18	741.72	4	2.85	6	--	--	0.828	0.319	0.003	0.677
	AE	760.06		3	2.85	7	0.00	1	0.899	0.324	--	0.677
	CE	763.03		3	5.82	7	2.97	1	0.562	--	0.214	0.785
	E	778.09		2	26.99	8	24.14	2	0.001	--	--	1.000
Low Negative	ACE	893.52	869.06	4	4.11	6	--	--	0.662	0.053	0.099	0.848
	AE	887.92		3	4.63	7	0.52	1	0.705	0.186	--	0.814
	CE	887.48		3	4.18	7	0.07	1	0.759	--	0.134	0.867
	E	889.51		2	12.33	8	8.22	2	0.137	--	--	1.000
Withdrawn/Inhibited	ACE	378.36	353.90	4	15.90	6	--	--	0.014	0.165	0.317	0.518
	AE	379.64		3	23.30	7	7.40	1	0.002	0.530	--	0.471
	CE	373.56		3	17.22	7	1.32	1	0.016	--	0.432	0.569
	E	460.81		2	110.58	8	94.68	2	0.000	--	--	1.000
Low Reactive	ACE	696.01	671.56	4	9.56	6	--	--	0.140	0.141	0.000	0.859
	AE	689.90		3	9.65	7	0.09	1	0.209	0.141	--	0.859
	CE	692.60		3	12.36	7	2.79	1	0.090	--	0.065	0.935
	E	688.36		2	14.23	8	4.66	2	0.076	--	--	1.000

Note: BIC = Bayesian information criterion; −2LL = −2 times the log likelihood; #free = number of parameters freely estimated; df = degrees of freedom; Δdf = change in degrees of freedom; Δχ2 = change in chi-square value from the full (ACE) model to reduced models; h² = additive genetic, c² = shared environment, and e² = nonshared environment standardized squared parameter estimates. In each set of models, the bolded model was the best-fitting.

Infant Temperament Profiles and Parenting

Next, we examined associations of the probability of belonging to a certain infant temperament profile to parent affect and parenting stress. Analyses used mixed models to account for the nonindependence of assessing twins within families. As shown in Table 9, associations were very modest; only 6 of the 48 estimates were significant at p < .05. At 6 months, infant probability of being in the low negative profile was associated with fathers reporting significantly lower negative affect and parenting stress. Fathers also reported higher parenting stress when their infant was in the withdrawn/inhibited profile at 6 months. At 12 months, for families with infants more likely to be in the typical profile, mothers reported lower negative affect. Infants in the withdrawn/inhibited profile had parents who reported significantly more negative affect for mothers and more stress for fathers (see Table 9).

Table 9.

Estimates from Multilevel Mixed Models to Determine Relation between Probability of Being Classified into Lab-TAB Observed Temperament Profiles at 6 and 12 Months and Parent Affect and Stress

Profile Classification Probability	Parent Characteristics
6 months	Mother Positive Affect	Mother Negative Affect	Mother Parenting Stress	Father Positive Affect	Father Negative Affect	Father Parenting Stress
Profile 1 / Typical	−.03	−.03	.01	.01	.01	.02
Profile 2 / Low Negative	.05†	.01	−.08†	.01	−.07**	−.15***
Profile 3 / Withdrawn/Inhibited	.003	.02	.06†	−.05†	.04	.09*
Profile 4 / Positive/Active	−.03	.01	.002	.02	.02	.04
12 months
Profile 1 / Typical	.01	−.06*	−.01	.01	−.03	−.05
Profile 2 / Low Negative	.01	−.03	−.06	.04	.01	−.05
Profile 3 / Withdrawn/Inhibited	.003	.05*	.05†	−.03	.03	.08*
Profile 4 / Low Reactive	−.02	.04†	.02	−.02	−.01	.01

Note:

^†p < .10,

^*p < .05,

^**p < .01,

^***p < .001. Regressions only within time point, not across time. Multilevel mixed models account for nonindependence of twins within family.

Discussion

We derived four temperament profiles during infancy using observations from the Lab-TAB. Previous research has identified between three and six temperament profiles in infancy and early childhood using multiple parent reported temperament dimensions. We found four profiles that were somewhat analogous across two ages in infancy: typical, low negative, and withdrawn/inhibited at both ages, a positive/active profile at 6 months, and a low reactive profile at 12 months. Measurements of infant temperament derived from the Lab-TAB are standardized within the entire sample, so we can compare levels of each behavior across profiles. The typical profile was most common in our sample, with 34% of six-month-olds and 44% of 12-month-olds being characterized by approximately average levels of each temperament dimension measured. The low negative profile showed a pattern of temperament dimension expression similar to the typical profile, but with slightly lower levels of anger and higher levels of fear and sadness at each age. Social fear, as an adaptive emotion, is observed around eight months of age (Izard & Malatesta, 1987), but then shows a developmental trajectory with decreasing levels in later infancy (16 months of age) (Braungart-Rieker, Hill-Soderlund, & Karrass, 2010; Brooker et al., 2013). Thus, the slightly higher levels of fear exhibited by infants in the low negative profile at 12 months of age, but not at six, is developmentally appropriate for infants otherwise low in distress type behaviors.

Infants in the withdrawn/inhibited profile exhibit higher levels of fear and sadness. This group comprised 19% and 15% of the sample at six and 12 months, respectively. Not only does the withdrawal behavior pattern mimic Kagan’s definition of behavioral inhibition, but the percentage of infants classified as withdrawn are also similar to the percentage of infants Kagan considered behaviorally inhibited (Coll, Kagan, & Reznick, 1984). Multiple other reports of infant behavioral inhibition (Biederman et al., 2001; Coll et al., 1984; Essex, Klein, Slattery, Goldsmith, & Kalin, 2010; Kagan, 1997) as well as work on infant temperament profiles (Beekman et al., 2015) find similar percentages of infants who express high withdrawal or inhibition behaviors. Notably, this group was also higher in activity level than other groups. This higher activity is also consistent with Kagan’s description of infants who are higher in motoric and crying behaviors later becoming inhibited (Kagan & Snidman, 1991).

Finally, the positive/active and low reactive profiles show a similar pattern at six and 12 months. This pattern is characterized by low to mid-range levels of anger and sadness at each age while the six month-olds had slightly higher levels of positive affect, interest, and activity level than the 12 month-olds. Of note, this profile shows the same pattern of elevation on the various dimensions at each age, yet with lower levels at 12 than six months. As regulatory systems come on-line in later infancy (Kopp, 1982), infants are better able to regulate extreme shifts in positive affect and activity level, and therefore may show more blunted responses to stimuli at 12 months of age.

Infant Temperamental Profile Stability

Profiles also showed mild to moderate stability across time, with profile stability highest for infants classified in the withdrawn/inhibited profile. Infants typically classified as behaviorally inhibited, or high on withdrawal behaviors such as fear, also tend to be inhibited later in life (Biederman et al., 2001; Essex et al., 2010). Though not conceptualized in the same way as fear or inhibition (Coll et al., 1984; Janson & Mathiesen, 2008), sadness is also a withdrawal behavior. Thus, the greater stability of the withdrawn/inhibited profile which is characterized by particularly high expressions of sadness during episodes which elicit distress could have been anticipated.

Unlike previous accounts of temperament profile stability, we found only modest stability in two profiles (typical, low negative), and no stability in the fourth profile. Such inconsistency in findings may indicate that observed expressions of temperamental behavior are: 1) not as stable as parents’ perceptions of temperament, or 2) are altered by environmental demands, reflecting patterns of temperamental behaviors that are also developmental in nature (Rothbart & Bates, 2006). Between six to 18 months of age, infants are continuously developing more advanced motor and cognitive behaviors that impact affective expression (Kopp, 1982; Planalp, Van Hulle, Gagne, et al., 2017). For example, pre-locomotor infants are not able to motorically seek out help from caregivers or external sources of support. As a result, crying may be as much a mode of communication as an indicator of distress. As the same infants become more ambulatory, they can use their mobility to more effectively regulate their distress, perhaps altering the behavioral profile expected based on previous observation.

Twin Similarity for Temperament Profiles

Genetic underpinnings are a common expectation for temperament (Buss & Plomin, 1975). We found that twin pair similarity was higher for MZ than DZ twins in almost all cases, with the strongest similarity in the withdrawn/inhibited profile. In addition, biometric models indicated only additive genetic and nonshared environment contributions to variation in the withdrawn/inhibited profile at each age; the best fitting model was the reduced AE model. This is, perhaps, unsurprising as negative aspects of temperament also show stronger genetic effects than positive aspects (Clifford, Lemery-Chalfant, & Goldsmith, 2015; Goldsmith et al., 1999; Planalp, Van Hulle, Lemery-Chalfant, et al., 2017). Notably, the fourth profile at each age (positive/active and low reactive) showed no evidence for genetic effects from either the ICCs or genetic models, where the “E only” reduced model fit the data best. As discussed previously, this profile also shows the least stability, and a pattern of less extreme behaviors between six and 12 months. One possibility is that this fourth profile is simply less valid than the other three. Alternatively, this profile may be most reflective of individual differences in learning and show most evidence of developmental transformations between six and 12 months (Kopp, 1989).

Parenting

Previous research has elucidated temperament typologies similar to ours using parent report, but only one study reported how parent characteristics predict child temperament (Beekman et al., 2015). Beekman et al. (2015) found that adoptive, but not birth parents’ personality related to infant temperament profiles. However, because adoptive parents reported on their own and the infants’ behaviors, reporter bias might have influenced the findings. In contrast, we compared observed temperament profiles to parent reports of their own affect and stress levels without compounding similarities by using independent infant and parent observations.

In our work, parenting stress was very modestly related to having an infant classified in the withdrawn/inhibited profile at each age. At six months, infants in the withdrawn/inhibited profile also showed higher activity level. If we had only examined parenting stress and activity level, we might erroneously draw a general conclusion that these two dimensions were related, when in fact it is only for children who were also higher in sadness that child temperament and parenting stress interrelate.

For infants in the low negative profile, parents reported lower parenting stress, though this effect was only significant at six months of age. Fathers also reported higher stress when infants were more likely to be classified in the withdrawn/inhibited profile at both ages. Fathers do tend to be involved more in active play behaviors with infants (Pleck, 2010). One component of parenting stress as assessed by the PSI is the parents’ connection with the infant. Conceivably, fathers find it more difficult (i.e., more stressful as a parent) to engage with an infant with higher withdrawal tendencies. Mothers also reported less negative affect at 12 months when their infants were in the typical profile and more negative affect with infants were in the withdrawn/inhibited profile. These findings do not diverge from published work on parenting and infant temperament. For example, in a sample of four-month-olds, infant temperamental negativity related to fathers’ parenting stress (Potapova, Gartstein, & Bridgett, 2014), and mothers reporting higher stress had infants lower in smiling and laughter (Bridgett et al., 2013). These relations are not limited to infancy, as infant emotional intensity and difficult temperament also predicts parenting stress for both mothers and fathers at 4 years of age (McBride, Schoppe, & Rane, 2002).

Notably, in our analyses and for each of aforementioned examples, we do not know whether infant characteristics lead to parenting, or vice versa. It is also possible that there is a heritable component of emotionality, such that parents of more negative infants also experience higher levels of negativity and stress. Nonetheless, by using a typology (profile) rather than dimensional approach, we garner a more integrated picture of the infant, and potentially a more complete picture of how parent and child characteristics interact within a family system as a whole.

Limitations and Conclusions

A limitation of any attempt to identify profiles or clusters of infant behavior is that the fit of any solution is always judged relevant to alternatives with the same dataset. We acknowledge that other solutions may be as useful as ours. Different, more optimal solutions could always be possible with additional variables, more participants, or different participants. For instance, Gartstein and colleagues (2017) identified three and five profiles for younger (3 to 8 months) and older infants (9 to 12 months), respectively, yet our data indicated a four-profile solution at six months and 12 months of age. The constructs chosen to be indicator variables of temperament profiles (in this case fear, anger, sadness, positivity, attention, and activity level), may alter or limit the number of profiles identified. Gartstein et al. (2017) used 14 scales from a parent report instrument, which may have led to them finding a clear five-profile solution in older infancy, as compared to our four-profile solution. Researchers who use different indicators will find different profiles. However, rather than viewing latent class analysis as an effort to identify illusive “essential” categories, we simply seek to identify profiles that will prove to be more or less useful in understanding temperament and its functional significance.

In contrast to existing research, we used observed measures of infant temperament and not parent report. Though we did not compare parent and observed measures, other groups have already identified temperament profiles using parent report. Although comparing profiles derived from both methodologies in the same children would be informative, our observational data approach adds to the literature on person-centered approaches to temperament.

We also used a sample of infant twins for our analyses. This allowed us to examine possible genetic effects on temperament profiles. We attempted to control for within family similarity by using independently observed data from the Lab-TAB, where trained coders did not score observations from both twins within the same family. Further, the profile derivation and subsequent analyses comparing profiles to parenting measures used mixed model techniques that account for family wide, interrelated data. However, it is possible that, in the case of families with twins parenting stress might be heightened; yet, the nature of the interactions between temperament and parenting affect and stress seem unlikely to differ. Nevertheless, to ensure that our findings relating to infant temperament profiles and relations with parenting are generalizable to a broader population, replication of our classes with singleton data could be useful.

Despite these limitations, this study adds to a growing literature on identifying infant temperament typologies rather than limiting analyses to single temperament dimensions. We add to existing literature by using observed infant behavior during structured episodes instead of more widely used parental report questionnaires. We also incorporate mother and father reports of their own affect and stress. Decades of research have focused on mothers’ contributions to infant development, yet less research has focused on fathers’ contributions (Pleck, 2010). Our focal argument is that studying multiple dimensions of infant temperament simultaneously is necessary to better characterize infants in the service of studying family processes. Similarly, including both caregivers in a household, where applicable, helps to better understand the complete family system. These more comprehensive approaches enhance our potential to understand how complex units within the family system, as well as multiple aspects of an infant’s temperament profile, interact to influence parenting and potentially later child outcomes.