Does the EDI Measure School Readiness in the Same Way Across Different Groups of Children?

Abstract

The present study investigates whether the Early Development Instrument (Offord & Janus, 1999) measures school readiness similarly across different groups of children. We employ ordinal logistic regression to investigate differential item functioning, a method of examining measurement bias. For 40,000 children, our analysis compares groups according to gender, English-as-a-second-language (ESL) status, and Aboriginal status. Our results indicate no systematic measurement differences regarding Aboriginal status and gender, except for 1 item on which boys are more likely than girls to be rated as physically aggressive by Kindergarten teachers. In contrast, ESL children systematically receive lower ratings on items of the language and communication domains—as expected by definition of ESL status—but not within the physical, social, and emotional domains. We discuss how our results fit with child development research and the purpose of the Early Development Instrument, thus supporting its validity.

Notes

¹Here, we use the term community to refer solely to the concept that is delineated in the British Columbia Atlas of Child Development (Kershaw, Irwin, Trafford, & Hertzman, 2005), which uses it synonymously with neighborhood.

²In BC, both the EDI and the Ministry of Education data included individual child information. Thus, children that were identified as ESL or Aboriginal in the EDI database but not in the Ministry database could be individually identified.

³The EDI is available (in English and French) at www.offordcentre.com/readiness/EDL_viewonly.html

⁴ The British Columbia Atlas of Child Development is available at www.earlylearning.ubc.ca

^aEffect sizes (Cohen's d) of .2, .5, and .8 are considered as small, medium, and large, respectively (Cohen, 1992).

^bSum of domain scores; 50-point scale.

^aEffect sizes of R ² < .035 are considered negligible, those between .035 and .070 moderate, and ones =.070 large (Jodoin & Gierl, 2001). Negligible effect sizes are in italics.

^bResponse options for these items are on a 3-point Likert scale: very good/good (10), average (5), poor/very poor (0), and don't know.

^cResponse options for these items are binary: yes (10), no (0), and don't know.

⁵Due to the absence of a statistical test, we cannot refer to this difference as statistically significant, even though the size implies practical significance.

⁶In the British Columbia Atlas of Child Development (Kershaw et al., 2005), quintiles are used for the reporting of results at the community and district levels.

⁷Currently, the Human Early Learning Partnership at the University of British Columbia, the organization that coordinates the EDI project in BC, is collaborating with numerous stakeholders toward developing an early childhood education tool that more clearly integrates Aboriginal values and their cultural diversity.