A Synthesis of Research on Reading from the National Institute of Child Health and Human Development

How the NICHD Research Program is Different

To appreciate fully the significance of the NICHD findings it helps to understand the level of scientific rigor used to guide the formation of conclusions from the research. Reid Lyon coordinates the parallel investigation of similar questions across several NICHD research centers. Under Lyon’s leadership, the researchers determine that the questions have been answered only when the findings replicate across researchers and settings. Findings with a high degree of replicability are finally considered incontrovertible findings and then form the basis for additional research questions. Funding is awarded the research centers through a competitive peer review process. A panel of researchers who are not competing for the research funds award the funds after evaluating competing proposals according to specific criteria. Each research study within the NICHD network must follow the most rigorous scientific procedures.

True scientific model. The NICHD studies do not embrace any a priori theory, but test all theories against one another at different points in time. In a true scientific paradigm, theories are tested by doing everything to try to prove the theory incorrect. This contrasts with the usual nature of research in education, where untested hypotheses are often presented as proved theories before any testing has occurred.

Long-term duration. The average length of a study has been 8 years, ranging in length from 3 years to 31 years. In these longitudinal studies, the growth of children from preschool through adulthood has been evaluated. Currently, several large-scale, 5-year longitudinal treatment intervention studies are underway. This longer-term design allows evaluation of the effects of different instructional variables on later reading performance.

Sampling procedures. The sampling procedures ensure that all subgroups in the population (all ethnic groups, a full range of IQ levels, and so on) are included in sufficient numbers to provide a window to the population as a whole and provide information regarding the relationship of reading disabilities to other variability in individuals such as IQ. To evaluate the relationship between IQ and reading disabilities, for example, the research subjects must proportionately sample different IQ bands. Most studies involve around 200 subjects representing variation within specified dimensions. Children who do not speak English have been excluded from the NICHD research samples to this point. After basic reading instruction issues have been resolved for teaching children with some knowledge of English, including bilingual children, the research questions will turn to treatment for children who do not know English and are beginning to learn it as a second language.

Researcher bias. Researcher bias is reduced by the sheer number of people involved in the NICHD program. For example, at only one NICHD-funded research center, the one at Yale University, the following researchers are involved: Jack Fletcher, David Francis, Rafael Kloorman, John Gore, John Halahan, Robert Constable, Leonard Katz, Barbara Foorman, Bonita Blachman, Dorothy Aram, Alvin Liberman, Ken Pugh, Michael Studdert-Kennedy, Donald Shankweiler, Karla Stuebing, Keith Stanovich, Linda Siegel, and Louisa Moats. In addition, researchers at the different NICHD centers communicate frequently regarding their findings, checking each other’s data and testing alternative explanations with additional studies.

Contrast with other educational research. The NICHD research program differs from much of the earlier research in its scientific rigor. Table 1 helps illustrate the contrast by summarizing several studies that reported conclusions that conflict with those of the NICHD. The studies in Table 1 are laudable for attempting to evaluate competing theories and were sometimes even two years in duration, quite long as educational studies go. Yet the studies are still too short in duration to evaluate the effects of the different treatments on the children’s actual ability to read with understanding. In nearly all of the studies in Table 1 the children never progressed far enough in their reading to use a measure of independent reading comprehension to evaluate their learning. The important question of how different approaches to beginning reading instruction ultimately impact authentic reading remains unanswered in these studies.

Many of the measures used to evaluate the children’s learning had no established validity as predictors of reading comprehension. For example, children who used multiple cueing systems or who said they valued understanding more than getting the words right, were given higher scores in many of the studies in Table 1. Whether or not this performance would correlate with later reading performance was not established at the time of the research.

With the NICHD research we now know that the values given the responses on these measures should have been reversed. What was considered desirable performance on miscue analyses actually indicates a poor comprehender, rather than a good comprehender. Children who are poor readers make greater use of two of the three cueing systems, syntax and semantics (context), than good readers. Good readers make greater use of the graphophonic cueing system, as indicated by the fact that they read fluently and accurately without rereading. Readers who get words right are better comprehenders than readers who guess using context to figure out words. Most likely the children who scored highest on these measures would become the poorest readers, based on NICHD studies of good and poor readers.

Even when the skills measured do predict better reading later, such as knowing the names of the letters, teaching children these skills does not necessarily guarantee that these children will be better readers later on. Though many of the studies in Table 1 were over two years duration, the time frame was still too short to see the nature of the impact of the instruction on reading comprehension.

Table 1. Research supporting conclusions that conflict with the NICHD research findings.

*N= number of subjects (Ss) in each treatment group.

In contrast, the NICHD longitudinal treatment studies now in progress are five years in duration and have already used reading comprehension measures to evalute instructional variables in the second year of the studies. In addition, the sample sizes are much larger in the NICHD research studies. For example, the kindergarten study by Foorman and her colleagues (in press) involved 260 kindergarten children. Their first- and second-grade study in eight Title I schools involved 375 subjects. Their special education study of children in the lower 25% involved 113 children with reading disabilities. The study of children in the lower 10% at the Florida Treatment Center involved 180 children (Torgesen et al., in press). The larger samples in the NICHD research included a full range of IQ levels, ethnic groups, and included lower income children. As Table 1 shows, the largest study reporting contradictory conclusions included only 100 subjects. Most of the studies involved much smaller samples.