Workshop on Understanding and Promoting Knowledge Accumulation in Education:

Tools and Strategies for Education Research

Day 1 – June 30, 2003

Remarks by Dr Ronald Ehrenberg

DR. RONALD EHRENBERG: Very early 1990s, I became very interested in issues relating to race, gender and ethnicity in American education. Since this was precipitated by actions that many school districts were taking to aggressively increase their hiring of under-represented minority faculty, even in the face of a declining pool of minorities seeking to enter careers in education and evidence that new minority teachers were failing the National Teacher Examination at a higher rate than new white teachers were.

Indeed, to save money and to create vacancies for new minority teachers, many school districts were providing financial incentives for their older, experienced and often white teachers to retire.

Now, the push to expand deployment of under-represented minority teachers was fueled by the belief that minority teachers were thought by many to be more effective teachers of minority students, because they could serve as role models, had more patient attitudes towards and expectations for and can provide more positive feedback to minority students, and, indeed, research by educational psychologists have supported some of these beliefs.

However, I was concerned that there had been little research conducted that focused on what the impacts of the match of teachers and students on race and ethnicity grounds was on the amount of student amount that students learned

There was also concern that trading off teacher race and ethnicity for experience and other dimensions of a teacher’s background, such as her educational preparation and academic ability might not prove to be in the best interests of either minority or other students. So I decided to conduct research on the topic.

Now, to do this required a data set that contains information on both the characteristics of students and their teachers. Ideally, the teacher data would include information on their educational background, their test scores and their experience. Ideally, the data would be individual in nature, so that students could be matched to specific teachers and would be longitudinal so that it could focus on the relationships between changes in students’ test scores during the time period and the characteristics of the teachers during this period.

Not surprisingly, as is often the case, no individual level data bases from a contemporary period provided the information that I needed for such a study. However, my student, Dominick Brewer, who is now Director of RAND Education, and I realized that if we were willing to make one crucial assumption, we could use school-level data from the classic mid-1960s Coleman Report to illustrate the nature of the tradeoffs that we were concerned about.

In particular, because the Coleman Report data were not longitudinal in nature, we needed to assume that differences in test-score performance between two grades in a school at a point in time were reasonable proxies for how much students would learn if they remained at the school for both grades.

With such an assumption, it would be possible for us to estimate what we called “synthetic game-score(?) equations” for white and black students at each school that related to synthetic game score to student, family, school, community and teacher characteristics.

Of primary interest to us would be the effect of the racial composition of teachers in a school and their verbal abilities on the synthetic game scores of students of each racial group.

Now, at the time we were contemplating our study, the Equal Employment Opportunity survey data base upon which the Coleman Report was based were available only from the National Archives. The data that we received from the National Archives was very poorly documented, and we began contacting researchers who were still alive who had been associated with the report - (laughter) - after all, it’s now 30 years later - to try to - better documentation.

Christopher Jenks(?), then at Northwestern, suggested that we contact Marshall Smith who was then Dean of the Stanford Education School, and fortunately Smith provided some useful information to us.

It turned out that although over 50,000 teachers were surveyed in the original EEO survey, the data that we received from the National Archives contained information on only 44,193 teachers and the data set came with a note that 2,000 teacher data were missing.

I was outraged that some of the data from what had arguably been the most important social science data set of its era could have been lost and that the data set was so inaccessible to researchers who saw merit in reanalyzing it.

At the time, I was a member of the Council of the Interuniversity Consortium for Political and Social Research, ICPSR, a national social-science data archives to which most academicians belonged. I know that Barbara Schneider described ICPSR a little bit for you in her presentation today.

Concern that such an incident should never occur again, that what was left of the EEO data be protected and be made available to researchers in a more understandable form, and, more generally, that an archive be established of data sets useful for research and educational issues, I encouraged ICPSR to expand its mission to include establishing a special data archive for educational data, and I was delighted that shortly after I left the ICPSR Council, ICPSR, together with the National Center for Education Statistics, established the International Archives for Education Data at ICPSR to facilitate research on educational issues.

Now, without going into the details of Brewer’s and my research and the methodological issues that related to it, we concluded that teachers’ race and verbal aptitude scores didn’t matter in the U.S. in the 1960s in the sense that they were associated with students’ synthetic game scores.

Higher teacher verbal scores were associated with higher synthetic game scores for both white and black students in a school, holding up a factor as constant. Put another way, teacher academic ability did matter.

In some specifications we found that a higher percentage of black teachers in a school was associated with lower synthetic game scores for white elementary and secondary school teachers and higher black student synthetic game scores only for black high-school students.

We stressed in our paper that what was true for the 1960s was not necessarily true for the 1990s, and that research needed to be conducted using more contemporary data bases.

We also stressed that other outcomes needed to be addressed, including drop-out rates and college-going rates.

Finally, we stressed that educational policy decisions need to be based on the accumulation of evidence from a variety of studies that expressed an issue using different data bases and different methodologies.

In a subsequent paper, Brewers, who is now at the University of Washington, and I were able to use the National Educational Longitudinal Survey of 1988 to revisit this question.

The advantages of the NELS data - was that it was proven by its(?) nature. It permitted us to match individual students with their classroom teachers. It came from a more contemporary period and it allowed us to extend the analysis to consider the matching of teachers not only by race but also by gender and ethnicity.

NELS ‘88 major disadvantage was that it contained no information on measures of future academic ability.

And we used data from the first two waves(?) of NELS which allowed us to study student game scores between the eighth and tenth grade and found that, for the most part, the match of teachers and students by race, gender, ethnicity did not effect how much students learned.

However, we did find evidence that in some cases a match influenced teacher’s subjective evaluation of students, which may influence students’ aspirations and the tracks on which they are put.

The final paper that I want to mention to you is a forthcoming paper by Thomas Deeds, an economist at Swarthmore, which is being published in the Review of Economics and Statistics. Deed’s data from Tennessee(?) Project Start classroom experience to revisit the issue.

The use of data from a random experiment was a distinct improvement on the earlier studies and Deeds finds evidence that assignment to own-race teachers increases the math and reading scores of both black and white students who took part in the Start experiment.

Put another way, relative to white teachers, black teachers helped black students, but hurt white students, and relative to black teachers, white teachers helped white students, but hurt black students.

While Deeds offers all the caveats that I have offered above about the need to focus on other outcomes and to see if the results can be generalized to other settings, his findings should stimulate us to think more about the need to train teachers to be effective with all population groups, regardless of their race or ethnicity.

I would conclude by noting that neither the Coleman Report data, the NELS ‘88 data nor the Start experiment data will collectively be ideal at analyzing the research question that was of concern to us.

However, their public availability made it possible for the research to be conducted.

Thank you.