Table 5.

Recommendations for research on the role of age of starting early intervention for children with autism spectrum disorder.

Area	Recommendations
Conceptual	(i) Planned analyses should be conceptually based
	(ii) Models will benefit from including both child-focused and environmental variables when considering the child variable of baseline age
	(iii) Conceptual models should take into consideration interaction of predictors
Participants	(i) Be aware of range of intellectual ability and proportion of severity levels present in the sample
	(ii) Include as broad a sampling as possible in terms of participant ability in cognition, play, language, and autism severity
	(iii) It would be helpful to have a more consistent definition of early intervention, distinguishing birth-to-three from preschool intervention
	(iv) School age (5 years and above) should not be considered early intervention
Measures	(i) When using standardized tests, consider value of age equivalent vs. standardized scores
	(ii) Explore ways to include children who cannot complete a standardized test
	(iii) Consider the constructs measured and separate out neuropsychological features such as language-based versus nonlanguage constructs and quotients
Approach to data analysis	(i) Explore distribution shape of continuous variables and adjust for skewness
	(ii) Testing of prediction relationships needs to move past zero-order correlations. Since starting age is an important theoretical predictor, lack of significant zero-order correlations may be bypassed for inclusion in further analysis because of the possibility of more complex relationships
	(iii) Multivariate approaches should control for shared variance among predictors
	(iv) Consider using statistical tests that are robust to small samples and nonparametric data (e.g., bootstrapping techniques) to minimize the possibility of type I and type II errors
	(v) Studies with large samples should consider more contemporary statistical approaches such as structural equation modeling in lieu of conducting multiple separate univariate and multivariate regression analyses
	(vi) Post hoc techniques for understanding the direction and magnitude of influence of age as predictor will be helpful