How can the EDI be used for research purposes?

For community reporting purposes, we recommend that it is used in the first or second year of formal schooling. However, in terms of research, it could be used outside of this time-frame. It very much depends on the population.

The best set of articles that refer to the EDI use both in research and at the community-level can be found in the special issue of the journal Early Education and Development (2007, 18 (3)). There are several research projects described in the papers there. The ones that Dr. Magdalena Janus is a co-author on are available in PDFs on our website (publications), and we have others available upon request.

Pilot studies – validity and reliability

There is no need to run the other test measures for a whole pilot group; it has never been done this way for any of the international sites. The rule of thumb for sample size is based on research design requirements and the number of instruments used – at least 5-10% of the full sample, with no fewer than 200 children. Unless you know that one region is very different in terms of population served (e.g., one has a large proportion of immigrants), you may want to focus your validity study on just one area, which would make it easier logistically. Our own “enriched sample” when we were starting with the EDI was about 20% of the total, with just over 2,100 children, and it was representative of six areas, but recruitment and data collection was carried out with the help of Statistics Canada.

Can I use only one component of the EDI in research?

Using just one component of the EDI, such as the social/emotional development of children, is not recommended. The EDI looks at the development of the child from a holistic point of view. For example, a child may have good language and cognitive skills, but if they are not doing well emotionally or socially they may not be successful in school. The same could be said for a child who gets along well with others and is mature emotionally but has poor language and cognitive skills. Success in all five domains is what contributes to the child’s ability to succeed in the school environment.

Why is the EDI not used on 3.5 year-old children?

It would not be ideal to use the EDI for 3.5 year olds as some items may be too difficult for their age. That said there have been a couple of studies where it has been used for children as young as this in a research project.  If you were planning a longitudinal assessment where you will be looking at the same group of children at age 5 or 6 then this would be fine.

Can the EDI be used to assess program effectiveness?

The EDI can be used to evaluate programs, but only within the appropriate research design (i.e., there would need to be a control group and data on other variables that may affect the outcomes). Please see the “How to use the EDI” section for more information on using the EDI for research purposes. The EDI Guide would also be a good resource.

What are the mean EDI scores?

It is appropriate – and a common practice – to collect the data categorically, and then convert it into a mean as if the data were continuous. Even though the items that make up a domain are categorical (or ordinal) the items were rescaled to a common metric and to ensure equal weighting the means of items are not used in the analyses. The total score for a domain is the best estimator that reflects a child’s level of ability for that domain. Work on the psychometric properties of the domains has been published using both Rasch Measurement Modeling and Classical Test Theory (publications). We do not endorse using the mean score as the only outcome reported. We recommend reporting vulnerability or the complete distribution of scores by domain along with the mean score.

Can I use a total EDI score in my analyses?

We do not endorse the use of a total EDI score. The five EDI domains are not cumulative in nature and explore different concepts. It is not uncommon that a child only has difficulties in one area/domain and does require intervention. However, by using the EDI total score you would mask these difficulties as it would not be reflected in a total EDI score.

Why is the lowest 10th percentile used as the cut-off?

The reason for using the lowest 10% is based on the existing research with instruments dealing with child development outcomes similar to the EDI, and the clinical standards for assuming that some deficiency would be a “problem”. A good example would be the Child Behaviour Checklist (CBCL). The scores in the lowest 3rd (or 2nd) percentile are customarily accepted as signifying a serious problem, while those within the lowest 5th percentile as a clinical problem. In the case of the CBCL, these were confirmed by comparison with referral status (i.e., whether the child was referred for a clinical treatment). Since the EDI was not developed as a diagnostic tool, it was not appropriate to compare the scores in the same way; however, the EDI is based on several existing instruments, including the CBCL, and therefore it was appropriate to use similar reasoning.

For the language and communication domains, we carried out some sensitivity/specificity analyses in concordance with direct language assessment. The EDI’s mandate is to identify areas of weaknesses in groups of children, and therefore, without the diagnostic accuracy, we felt that the 5th percentile boundary would be too small. This was the decision of one of the co-authors, Dr. Offord, a child psychiatrist, based on his long clinical experience and clinical judgment.

The 10th percentile was therefore chosen as close enough to capture children who were struggling, but not only those who were doing so visibly to have already been identified. Using the 10th percentile has good external validity, as falling within the lowest 10th percentile on the EDI is predicted by many variables associated with negative outcomes in the literature.

Why aren’t children with special needs included in the Normative II dataset?

Any normative data have to be normed on a typically developing population, as otherwise the perception of normal development will be skewed. In extreme cases, this might lead to acceptance of suboptimal developmental status as adequate. This follows the World Health Organization (WHO) recommendations. For example, when the WHO produces growth norms, children with stunting are not included in the population on which the percentiles are normed.

Why report on “Vulnerability”?

Data are collected for whole populations of children. The focus of the interpretation is on universal action: prevention reaching all children. Reporting on “Vulnerability” is based on the premise that small modification of risk for large numbers is more effective in producing change than large modifications for small numbers. “Vulnerability” provides a glimpse into the groups of children who are vulnerable to problems in later childhood — it casts a wide net, including all children who may benefit from universal programs. Children with more risks have more complex problems and require more interventions. Reporting on “Vulnerability” rather than a total score ensures that children’s vulnerabilities are not masked by excellence in another area.

The predictive validity of “Vulnerability” is strong, as data from the Toronto District School Board has shown 40-50% of children falling in the “Vulnerability” category have difficulties in Grade 6, and data from BC shows 34% of these children have difficulties in Grade 4.

The “Vulnerability” group represents children for whom cost-effective, universal preventive programs are likely to make a difference. Reporting on these children reflects the fundamental premise on which the EDI’s concept has been built. It reflects the population that we are most likely to shift without costly and intensive interventions.

Why doesn’t the EDI use confidence intervals?

Confidence intervals are typically used when using a random sample of a larger population in order to estimate a range of values likely to contain the population parameter of interest. The EDI is a population-level measure of children’s developmental health. Accordingly, statistics such as mean EDI domain scores are developed from an entire population of children and represent the population mean. Creating an estimate of the range of values of the population mean is unnecessary as our data are from a total population, not a sample.