Last week I received an email from Sunil Iyengar of the National Endownment for the Arts responding to Nancy Kaplan’s critique (published here on if:book) of the NEA’s handling of literacy data in its report “To Read or Not to Read.” I’m reproducing the letter followed by Nancy’s response.
Sunil Iyengar:
The National Endowment for the Arts welcomes a “careful and responsible” reading of the report, To Read or Not To Read, and the data used to generate it. Unfortunately, Nancy Kaplan’s critique (11/30/07) misconstrues the NEA’s presentation of Department of Education test data as a “distortion,” although all of the report’s charts are clearly and accurately labeled.
For example, in Charts 5A to 5D of the full report, the reader is invited to view long-term trends in the average reading score of students at ages 9, 13, and 17. The charts show test scores from 1984 through 2004. Why did we choose that interval? Simply because most of the trend data in the preceding chapters–starting with the NEA’s own study data featured in Chapter One–cover the same 20-year period. For the sake of consistency, Charts 5A to 5D refer to those years.
Dr. Kaplan notes that the Department of Education’s database contains reading score trends from 1971 onward. The NEA report also emphasizes this fact, in several places. In 2004, the report observes, the average reading score for 17-year-olds dipped back to where it was in 1971. “For more than 30 years…17-year-olds have not sustained improvements in reading scores,” the report states on p. 57. Nine-year-olds, by contrast, scored significantly higher in 2004 than in 1971.
Further, unlike the chart in Dr. Kaplan’s critique, the NEA’s Charts 5A to 5D explain that the “test years occurred at irregular intervals,” and each test year from 1984 to 2004 is provided. Also omitted from the critique’s reproduction are labels for the charts’ vertical axes, which provide 5-point rather than the 10-point intervals used by the Department of Education chart. Again, there is no mystery here. Five-point intervals were chosen to make the trends easier to read.
Dr. Kaplan makes another mistake in her analysis. She suggests that the NEA report is wrong to draw attention to declines in the average reading score of adult Americans of virtually every education level, and an overall decline in the percentage of adult readers who are proficient. But the Department of Education itself records these declines. In their separate reports, the NEA and the Department of Education each acknowledge that the average reading score of adults has remained unchanged. That’s because from 1992 to 2003, the percentage of adults with postsecondary education increased and the percentage who did not finish high school decreased. “After all,” the NEA report notes, “compared with adults who do not complete high school, adults with postsecondary education tend to attain higher prose scores.” Yet this fact in no way invalidates the finding that average reading scores and proficiency levels are declining even at the highest education levels.
“There is little evidence of an actual decline in literacy rates or proficiency,” Dr. Kaplan concludes. We respectfully disagree.
Sunil Iyengar
Director, Research & Analysis
National Endowment for the Arts
Nancy Kaplan:
I appreciate Mr. Iyengar’s engagement with issues at the level of data and am happy to acknowledge that the NEA’s report includes a single sentence on pages 55-56 with the crucial concession that over the entire period for which we have data, the average scale scores of 17 year-olds have not changed: “By 2004, the average scale score had retreated to 285, virtually the same score as in 1971, though not shown in the chart.” I will even concede the accuracy of the following sentence: “For more than 30 years, in other words, 17year-olds have not sustained improvements in reading scores” [emphasis in the original]. What the report fails to note or account for, however, is that there actually was a period of statistically significant improvement in scores for 17 year-olds from 1971 to 1984. Although I did not mention it in my original critique, the report handles data from 13 year-olds in the same way: “the scores for 13-year-olds have remained largely flat from 1984-2004, with no significant change between the 2004 average score and the scores from the preceding seven test years. Although not apparent from the chart, the 2004 score does represent a significant improvement over the 1971 average – ?a four-point increase” (p. 56).
In other words, a completely accurate and honest assessment of the data shows that reading proficiency among 17 year-olds has fluctuated over the past 30 years, but has not declined over that entire period. At the same time, reading proficiency among 9 year-olds and 13 year-olds has improved significantly. Why does the NEA not state the case in the simple, accurate and complete way I have just written? The answer Mr. Iyengar proffers is consistency, but that response may be a bit disingenuous.
Plenty of graphs in the NEA report show a variety of time periods, so there is at best a weak rationale for choosing 1984 as the starting point for the graphs in question. Consistency, in this case, is surely less important than accuracy and completeness. Given the inferences the report draws from the data, then, it is more likely that the sample of data the NEA used in its representations was chosen precisely because, as Mr. Iyengar admits, that sample would make “the trends easier to read.” My point is that the “trends” the report wants to foreground are not the only trends in the data: truncating the data set makes other, equally important trends literally invisible. A single sentence in the middle of a paragraph cannot excuse the act of erasure here. As both Edward Tufte (The Visual Display of Quantitative Information) and Jacques Bertin (Semiology of Graphics), the two most prominent authorities on graphical representations of data, demonstrate in their seminal works on the subject, selective representation of data constitutes distortion of that data.
Similarly, labels attached to a graph, even when they state that the tests occurred at irregular intervals, do not substitute for representing the irregularity of the intervals in the graph itself (again, see Tufte and Bertin). To do otherwise is to turn disinterested analysis into polemic. “Regularizing” the intervals in the graphic representation distorts the data.
The NEA report wants us to focus on a possible correlation between choosing to read books in one’s leisure time, reading proficiency, and a host of worthy social and civic activities. Fine. But if the reading scores of 17 year-olds improved from 1971 to 1984 but there is no evidence that during the period of improvement these youngsters were reading more, the case the NEA is trying to build becomes shaky at best. Similarly, the reading scores of 13 year-olds improved from 1971 to 1984 but “have remained largely flat from 1984-2004 ….” Yet during that same period, the NEA report claims, leisure reading among 13 year-olds was declining. So what exactly is the hypothesis here -? that sometimes declines in leisure reading correlate with declines in reading proficiency but sometimes such a decline is not accompanied by a decline in reading proficiency? I’m skeptical.
My critique is aimed at the management of data (rather than the a-historical definition of reading the NEA employs, a somewhat richer and more potent issue joined by Matthew Kirschenbaum and others) because I believe that a crucial component of contemporary literacy, in its most capacious sense, includes the ability to understand the relationships between claims, evidence and the warrants for that evidence. The NEA’s data need to be read with great care and its argument held to a high scientific standard lest we promulgate worthless or wasteful public policy based on weak research.
I am a humanist by training and so have come to my appreciation of quantitative studies rather late in my intellectual life. I cannot claim to have a deep understanding of statistics, yet I know what “confounding factors” are. When the NEA report chooses to claim that the reading proficiency of adults is declining while at the same time ignoring the NCES explanation of the statistical paradox that explains the data, it is difficult to avoid the conclusion that the report’s authors are not engaging in a disinterested (that is, dispassionate) exploration of what we can know about the state of literacy in America today but are instead cherry-picking the elements that best suit the case they want to make.
Nancy Kaplan, Executive Director
School of Information Arts and Technologies
University of Baltimore
if:book
A Project of the Institute for the Future of the Book
To dispense with Nancy Kaplan’s more facile errors first: she is simply wrong when she says the NEA report mentioned the 1971 NCES data only once, in a “single sentence in the middle of a paragraph.” The fact that the NAEP long-term trend assessment has been conducted routinely since 1971 is noted four separate times on pp. 56-57 of the full report.
Second, Dr. Kaplan ignores the last paragraph of my earlier reply. She does not seem to understand that NCES’ explanation of the “statistical paradox” refers solely to the reason why the overall average reading score for adults failed to change between 1992 and 2003: because a higher percentage of adults gained a college education over that period and a lower percentage did not complete high school. The NEA duly reports this explanation, which does not apply to the following two statements in our report: “American adults of virtually all education levels are reading less well than in the previous decade” and “the decline in the percentage of Proficient readers is occurring at the highest educational levels.”
In other words, if you look at the average reading score of adults between 1992 and compare it with the score in 2003, you will find no change, for the reason stated above. But if you review average reading scores of adults by education level, the picture is quite different–and certainly worth reporting–as the NEA and NCES independently realized.
I appreciate Dr. Kaplan’s skepticism as a hallmark of scientific inquiry, but I am baffled by her use of the term “disingenuous” to characterize our honest engagement with her questions. She is correct in stating that a variety of time periods and ranges appear in this report, which draws from multiple large studies. Yet in our discussions of voluntary reading and reading proficiency, we made every attempt to stay within the most recent 20- or 25-year timeframe, again because the original NEA study–conducted with the U.S. Census Bureau since 1982–showed steep declines in literary and book reading in young adults. This concern largely impelled our follow-up report, To Read or Not To Read.
But Dr. Kaplan’s critique raises a larger philosophical issue, which readers of the data must weigh seriously. As a nation, should we be content with the mediocre reading ability of our youth? Put another way, is the report wrong to highlight the decline in reading proficiency among 17-year-olds from 1984 (the first year of a statistically significant gain from 1971 test results) and 2004? Is the report wrong to show that 13-year-old reading scores have remained flat from 1984-2004, though slightly higher than in 1971? Or that 9-year-olds, of which more than 50% have read for pleasure “almost every day” from 1984-2004, saw significant gains in their average reading score over that period?
Nowhere does the report pretend there is a one-to-one correspondence between reading frequency and other outcomes. Rather, the results are shown responsibly, with voluntary reading as a significant contributor to reading proficiency. Even if they never have read our report, I doubt many educators would dispute this central point.
Finally, our staff is well-acquainted with the teachings of Edward Tufte, and was careful throughout this report not to violate his principles of visual data presentation.
Yours sincerely,
Sunil
Sunil Iyengar
Director, Research & Analysis
National Endowment for the Arts