It is impossible to begin this commentary without expressing appreciation that Cook et al1 explore how infant temperament assessment may help identify infants at risk for childhood behavioral problems. The authors report that infants with a severely unsettled profile of regulatory behavior have “staggeringly” high odds to have mental health difficulties at age 10 years. Developmental researchers and epidemiologists rarely report large effect sizes these days. Most common exposures such as parenting, diet, environmental toxins, and genetic variants have small or very small effects. The authors thus raise fascinating questions about infant temperament and the study’s methodology. Why might the predictive value of difficult temperament, 40 years after the seminal work of Thomas and Chess,2 be so much higher than ever observed before?
Cook et al1 followed 1002 mother-child pairs from age 12 months to 5 years and 871 of these pairs to 11 years. The study has distinct strengths: a population base, a good baseline response rate (82% of approached), and a reasonable 10-year follow-up (50%). At baseline, mothers completed 6 items on infant regulation (excessive crying, mood swings, temper tantrums, feeding and sleep problems, and global temperament), and at follow-up mothers filled out the Strength and Difficulties Questionnaire. The authors used a latent class analysis and selected a 5-class solution to group children into a settled (37%), tantrum (21%), sleep problems (25%), moderately (13%), and severely unsettled (3%) profile. Infants with a severely unsettled profile were much more likely (odds ratio 10.4; absolute risk ∼30%) to be scored in the clinical range of the Strength and Difficulties Questionnaire than those with a settled profile by their mothers 10 years later.
The authors use no established infant regulation scale. The authors refer to Fullard et al,3 who developed the Toddler Temperament Scale, but this instrument has 97 items. Today’s most used instrument, Rothbart and co-workers’4 Infant Behavior Questionnaire, Revised, has 91 items in the short version and 36 items in the very short version. Interestingly, the unsettled children identified by Cook et al1 are similar to Thomas and Chess’s2 “difficult” children, which are high in irregularity, withdrawal, negative mood, intensity of reaction, and are unadaptable to change.
So why could the effect estimates reported by Cook et al1 be so high? First, there is the seeming simplicity of infant behavior, which might be closer to the unadorned foundation of personality5; infant temper tantrums and unsettled behavior may be more predictive than later problems. Possibly, but in truth it is notoriously difficult to measure infant behavior reliably. Second, unsettled infants represent the most severe problem group on what, I would argue, is probably the extreme on 1 broad dimension. The authors thus study an extreme category in relation to a clinical category of an unspecific outcome: total problems, which cut across all disorder liabilities. Categorization of extremes on broad dimensions certainly helps find strong prospective associations, although it will make it difficult to understand the etiology and design interventions. Third, the authors asked the mothers to report infant temperament and child outcome. This is problematic because it introduces common method variance.6 Using the same method for a subjective exposure and outcome will have biased the association because answer patterns, informant cognition, or temperament certainly influenced the response. More than 50% of an association in developmental psychology can be accounted for by common method variance,7 certainly if researchers link traits that are not self-rated but rated by an informant. This is not to say that parents simply projected their own temperament in their ratings, but what is measured is the consistency in the eye of the beholder.5 Finally, the choice of statistical analysis for a prediction paper is unusual. The small number of items does not allow a factor analysis, so Cook et al1 assume that infant temperament is not continuously distributed and distinct groups of children should be identified. However, a 5-latent class solution based on 6 single-item indicators reflecting different subdomains is unlikely to be replicated and fit other data. Why not simply select the relevant items, standardize, and add them up? Equal weights would have had the greatest robustness in prediction.8
Cook et al1 raise important issues for the field of infant temperament. They suggest that predictive validity is key. Their temperamental measure is brief and basic, but if it validly predicted child and adolescent outcomes, it should be seriously considered. Soon this Australian cohort will have self-reported data to test this hypothesis. Also, the authors remind us that emotional dysregulation is not only a salient early predictor of poor child development but a characteristic of many psychiatric disorders.9 In small studies, large effect sizes often reflect chance; in large studies, methodological choices. Pediatricians should thus not take the risk estimates, which Cook et al1 present for children with an unsettled profile, at face value. However, the article affirms that mothers with unregulated children seeking help need careful counseling and advice.
Opinions expressed in these commentaries are those of the author and not necessarily those of the American Academy of Pediatrics or its Committees.
FUNDING: No external funding.
COMPANION PAPER: A companion to this article can be found online at www.pediatrics.org/cgi/doi/10.1542/peds.2018-0977.
POTENTIAL CONFLICT OF INTEREST: The author has indicated he has no potential conflicts of interest to disclose.
FINANCIAL DISCLOSURE: The author has indicated he has no financial relationships relevant to this article to disclose.