New Educational Foundations, Volume 1, Number 1
Access this review in .pdf format
The Rot Festers:
Another National Research Council Report on Testing
Hout, M., & Elliott, S.W. (2011). Incentives and test-based accountability in education. Committee on Incentives and Test-Based Accountability in Public Education. Board on Testing and Assessment, Division of Behavioral and Social Sciences and Education, National Research Council. Washington, DC: The National Academies Press.
reviewed by Richard P. Phelps
In research organizations that have been “captured” by vested interests, the scholars who receive the most attention, praise, and reward are not those who conduct the most accurate or highest quality research, but those who produce results that best advance the interests of the group. Those who produce results that do not advance the interests of the group may be shunned and ostracized, even if their work is well-done and accurate.
The prevailing view among the vested interests in education does not oppose all standardized testing; it opposes testing with consequences based on the results that is also “externally administered”—i.e., testing that can be used to make judgments of educators but is out of educators’ direct control. The external entity may be a higher level of government, such as the state in the case of state graduation exams, or a non-governmental entity, such as the College Board or ACT in the case of college entrance exams.
One can easily spot the moment vested interests “captured” the National Research Council’s Board on Testing and Assessment (BOTA). BOTA was headed in the 1980s by a scholar with little background or expertise in testing (Wise, 1998). Perhaps not knowing who to trust at first, she put her full faith, and that of the NRC, behind the anti-high-stakes testing point of view that had come to dominate graduate schools of education. Proof of that conversion came when the NRC accepted a challenge from the U.S. Department of Labor to evaluate the predictive validity of the General Aptitude Test Battery (GATB) for use in unemployment centers throughout the country.
Fairness in Employment Testing, 1989
From the 1960s to the 1990s, the field of personnel psychology (a.k.a. industrial-organizational psychology) produced an impressive body of technically advanced research on the costs and benefits of testing for personnel selection. Thousands (yes, thousands) of empirical studies were conducted in the United States alone, demonstrating that a fairly general aptitude or achievement test is the best single predictor of performance for the overwhelming majority of jobs, better than all other factors that employers generally used in hiring. The estimated net benefits of using tests for personnel screening were huge, with costs minuscule and benefits enormous.
In the late 1980s, the U.S. Department of Labor considered providing the federal government’s GATB, which was used for hiring in federal jobs, to local employment offices for use in hiring outside the federal government. The test would have been made available to job applicants who wished to take it, and test results would have been made available to employers who wished to review them.
The Labor Department asked the Board on Testing and Assessment at the National Research Council to review the question. Their report is extraordinary. In the face of overwhelming evidence to the contrary, the Board declared the following: there was only negligible evidence to support the predictive power of the GATB, and tests in general provided no benefits in personnel selection. Their conclusions were reached through tortuous illogic and contradiction, and a judicious selection of both committee members and research sources (see Phelps, 1999).
For example, not one of the hundreds of academic psychologists who studied personnel selection was invited to participate in writing the report, whereas several education professors who were well-known opponents of high-stakes testing were. Many of the world’s most-respected personnel and GATB testing experts were appointed to a “Liaison Committee”, but it was never consulted; their names, however, were then published in the final report, as if to imply they approved of the report.
They did not. Members of the Liaison Committee accused the NRC of choosing deliberately a committee they knew would be hostile toward the GATB research.
Moreover, only one of the thousands of empirical studies on personnel selection was discussed. In the face of thousands of predictive validity studies on general aptitude tests in employment, the study committee wrote: “very slim empirical foundation”, “the empirical evidence is slight”, “fragmentary confirming evidence”, “very little evidence”, “no well-developed body of evidence”, and “primitive state of knowledge.”
The Board dismissed the benefits of hiring better qualified applicants for jobs by arguing that if an applicant were rejected for one job, the applicant would simply find another somewhere else in the labor market, as all are employed somewhere. (No matter that the other job might be less well-paid, in an undesirable field or location, part time, temporary, or even nonexistent.)
In the view of the report, “unemployment is a job.” The Board continued with the astounding contradiction that, whereas selection (and allocation) effects should be considered nonexistent because all jobs can be considered equivalent, general tests, like the GATB, cannot be any good as predictors because these tests do not account for the unique character of every job.
Constants on NRC testing study committees for the past quarter century have been the multiple participation of members of the federally-funded Center for Research on Educational Standards and Student Testing (CRESST), headquartered at UCLA, and members of an even more radical (anti-) testing research center at Boston College. Committee memberships are then rounded out with scholars known in advance to support CRESST biases and a few others with recognizable names and ideological sympathies, but little familiarity with the study topic. The many scholars who disagree with CRESST’s point of view are neither invited to participate nor cited in the study reports.
High Stakes, 1999
The most revealing aspect of the National Research Council’s 1999 report, High stakes: Testing for tracking, promotion, and graduation (Heubert & Hauser) is its choice of source material. Sources were included that buttressed the views of the BOTA and hundreds of sources that did not were ignored. The majority of citations went to CRESST research and CRESST researchers. At the time, NRC’s Board was chaired by a CRESST director. The “Committee for Appropriate Test Use”, the entity responsible for the particular study, included three CRESST grandees and one from Boston College.
With huge resources at its disposal (a budget of over $1 million), the NRC Board minimized its research effort. On issue after issue, it threw its lot in with a single or a single group of researchers. The chapter on tracking is really about the work of just one person (UCLA’s Jeannie Oakes). The counterevidence and counterarguments on that issue are kept completely hidden from the reader. The early childhood, readiness testing, and promotion and retention sections also feature only one person’s point of view (that of Lorrie Shepard of CRESST). Chapter 10 cites only three sources (an earlier NRC report and Shepard and Linn of CRESST). Chapter 11 cites, essentially, only two sources, George Madaus and Walt Haney of Boston College. Two thirds of the citations in the report refer to less than a dozen research sources.
For a book on a psychometric topic, the NRC report strangely ignores psychology research. Only ten citations out of 400 come from psychology journals, and they pertain only to a discussion of assessment standards and theoretical concepts of validity. The report avoids, in its entirety, the huge mass of accumulated empirical evidence on high-stakes selection from psychology journals. The report refers exclusively to research in education journals and reports and, even then, only the work of a small group.
The opinions of the general public are dismissed just as effortlessly. The report (pp. 44–45) acknowledges the high level of public support for high-stakes, but discounts it thusly:
“Despite some evidence that the public would accept some of the potential tradeoffs, it seems reasonable to assume that most people are unaware of the full range of negative consequences related to . . . high-stakes test use. Moreover, it seems certain that few people are aware of limits on the information that tests provide. No survey questions, for example, have asked how much measurement error is acceptable when tests are used to make high-stakes decisions about individual students. The support for testing expressed in polls might decline if the public understood these things.”
Then again, it might not. Almost all adults are experienced former students. It so happens that they know something about school.
High Stakes includes over 40 recommendations. With some exceptions, any one of them taken alone seems reasonable. Taken together, they would impose a burden on the states that none could feasibly meet. The report even floats a proposal to require that tests be pre-tested, before they can be used for high-stakes purposes, using a new, very general standard of predictive validity. Because testing proponents argue that high-stakes tests promote more learning or better employment, the NRC Board argues that we should hold off certifying the use of any particular high-stakes test until it can be proven that, over time, the test does increase learning (say, in college) and improve employment outcomes. It would take years, of course, to conduct such an experiment, even if the experiment were feasible. But, of course, it is not. One cannot test the effects of high-stakes tests when the stakes are not high as, presumably, they would not be during the life of the experiment.
High stakes was released at a propitious time, just before the debate over and design of the No Child Left Behind (NCLB) Act. For those who regarded the National Research Council’s work to be objective and trustworthy, it would serve as a caution, and nothing more. A century’s worth of program evaluations and experimental research on the optimal design of high-stakes test-based accountability systems was ignored, relegated to an information abyss. When the nation needed the information most and was most ready to use it, the National Research Council suppressed it.
In response to the NRC’s deliberate neglect of the research literature, I began to study it myself. As I lacked the NRC’s considerable resources, it took me some time—a decade, as it turned out—to reach a satisfactory stage of completion. I hedge on the word “completion” because I do not believe it possible for one individual to collect all the studies in this enormous research literature …that CRESST officials claim does not exist.
To date, I have read over 3,000 studies, found about a third of them to be appropriate for inclusion in a summary of qualitative studies and meta-analyses of quantitative and survey studies. Most had been available to the NRC study group as well, but were implied not to exist. A summary of the study is published in the International Journal of Testing (Phelps, 2012). Source lists can be found here:
Perhaps not surprisingly, a review of a great expanse of the research literature, rather than just the selective, tiny bit covered by the NRC report, leads to quite different conclusions and policy recommendations.
Common Standards for K–12 Education? 2008
Almost two decades ago, while working at the U.S. General Accounting Office (GAO, now called the Government Accountability Office), I managed a study to estimate the extent and cost of standardized testing in the United States. At the time, then-president George H. W. Bush had proposed a national testing program, and the U.S. Congress wanted to know how much it might cost and the effect it might have on then-current state and local testing programs.
On every quality indicator (e.g., survey response rates, fact-checking) the study exceeded GAO norms. A Who’s Who of notables in the evaluation, statistical, and psychometric worlds reviewed various aspects of the study. Nothing like it in quality or scale had been done before—it included details from all 48 states with testing programs and from a representative sample of over 500 U.S. school districts. One would think the education research community would have been interested in the results (U.S. General Accounting Office, 1993).
I left the GAO for other employment before the report was actually released, however, and, apparently, the pressure to suppress the report and its findings (essentially, that standardized testing is not that burdensome and does not cost that much) descended even before it was released. Over the ensuing months, I became gradually aware of more efforts to suppress the report’s findings. Panels were held at CRESST conferences—panels to which I was not invited— eviscerating it and suggesting that better studies were needed. The characterizations of the GAO report were completely false: the critics claimed that information was left out that, in fact, was not, and that information was included that, in fact, was not. But, reasonable people, allowed to hear only one version of the story, believed it, and the GAO report, along with the most thorough and detailed data base on testing practices ever developed, faded into obscurity.
In its place, other reports were written and presented at conferences, and articles published in mainstream education journals, purporting to show that standardized tests cost an enormous amount and were overwhelming school schedules in their volume. The studies were based on tiny samples, a single field trial in a few schools, a few telephone calls, one state, or, in some cases, the facts were just made up. The cost studies among them that actually used some data for evidence tended to heap all sorts of non-test activities into the basket and call them costs of tests.
I contacted the researchers making the erroneous claims and the CRESST directors by email, postal letter, and telephone. In a few cases, I received assurances, first, that the matter would be looked into—it was not—and, second, that an erratum would be published in the CRESST newsletter—it never was.
I submitted articles based on the GAO study to mainstream education journals and they were rejected for outlandish and picayune reasons, or because "everyone knows" that the GAO report was flawed.
Ultimately, a summary of the GAO report won a national prize and was published in a finance journal (Phelps, 2000, Winter). I suspect, however, that if the GAO report had arrived at "correct" conclusions (i.e., that standardized tests are enormously expensive and otherwise bad) any article derived from it could easily have been published several years earlier in most any mainstream education journal.
One would think that the assault on the GAO study might have ended in the 1990s, given how successful it was. But, perhaps, the report’s quality or the GAO name is so durable that education insiders feel the need to condemn it even fifteen years later, as they have in the National Research Council report Common Standards for K–12 Education?
To my observation, the CRESST and NRC preferred practice for information suppression is to ignore or declare nonexistent any research that contradicts theirs. There are several advantages to this practice, the “dismissive review” (see Phelps, 2009):
first, it is easier to win a debate with no apparent opponent;
second, declaring information nonexistent discourages efforts to look for it;
third, because it is non-confrontational, it seems benign and not antagonistic; and
fourth, there is plausible deniability, i.e., one can simply claim that one did not know about the other research.
When only one side gets to talk, of course, it can say pretty much anything it pleases. With no counterpoint apparent, “facts” can be made up out of thin air, with no evidence required. Solid research supportive of opposing viewpoints is simply ignored, as if it did not exist. It is not mentioned to reporters, it is not cited in footnotes or reference lists. It is treated as if it was never done.
Dismissive reviews are not credible to outsiders, however, when contradictory research is widely known to exist. Thus, the research that remains—that which cannot credibly be dismissed as nonexistent—must, instead, be discredited. In such cases, the preference for dismissive reviews must be set aside in favor of an alternate strategy: misrepresent the disliked study and/or impugn the motives or character of its author.
And, so it has been with the GAO report on testing costs. The GAO manifests a prominent profile and a reputation that is not easily demeaned. Nonetheless, the federally-funded research center CRESST has worked tirelessly for two decades now to do exactly that. And, in that quest, CRESST has had a distinct advantage: it is mandated and funded to disseminate its findings, whereas the GAO is not. Once a GAO report is released and a GAO official testifies to its only client—the U.S. Congress—no further agency effort promotes the work. By contrast, CRESST’s mission and funding include promotion of its work through marketing and conferences.
This 2008 NRC report, released fifteen years after the GAO report on testing costs, asserts, again, that the GAO report left something out and so underestimated the cost of testing. And, again, the assertion is false. This time, the NRC accused the GAO of neglecting to consider the cost of standard-setting during test development; in fact, this cost was fully counted in the GAO estimate.
But, yet again, claiming a void in others’ calculations is used as an excuse to bulk up their own cost estimates massively. Here are just a few ways that the NRC report, Common Standards for K–12 Education?, overestimates the cost of testing:
• One-time-only start-up costs—e.g., standard setting—are counted as annual recurring costs.
• Educator travel and lodging expenses for serving on standard-setting and other test development panels are counted twice, both as direct educator expenses and in the budget of the state education agency (which, in fact, reimburses the educators for these expenses).
• The full duration of all testing activities at a school—said to be 3–5 days—is allotted to each and every educator participating. So, for example, the time of a fifth grade teacher who administers a one-hour math exam on Tuesday of testing week, and who otherwise teaches regular class that week, is counted as if s/he were involved in administering each and every exam in every subject area and at every grade level throughout the entire 3–5 days. Moreover, the time of each and every teacher in the school is counted as if each and every teacher is present in each and every testing room for all subject areas and grade levels. By this method, the NRC overestimates the amount of educator time spent directly administering tests about twenty-fold.
Another way of looking at it is to ignore the fact that a school administers a series of one-hour tests across the tested subject areas and grade levels over the span of 3–5 days but, instead, assume that all classes in all subject areas and grade levels are sitting for 3–5 days doing nothing but taking 3–5-day-long exams, which, in fact, is not what happens.
• The NRC calculates the number of teachers involved by using a federally-estimated average pupil-teacher ratio, rather than an average class size estimate. Pupil-teacher ratios underestimate class sizes because they include the time of teachers when they are not teaching. By this method, the NRC overestimates the number of teachers involved in directly administering tests by another 50%.
• The NRC counts all teachers in a school, even though only those in certain grade levels and subject areas are involved in testing—usually amounting to fewer than half a school’s teachers. By this method, the NRC overestimates the number of teachers involved in directly administering tests by another 50% or more.
• In calculating “data administration costs” of processing test data in school districts and states, the NRC classifies all who work in these offices as “management, business, and financial” professionals who make $90,000/year. Anyone who has worked in state and local government data processing departments knows that this would grossly overestimate the real wages of the majority of these employees who, essentially, work as clerical and, oftentimes, contingent staff.
• The NRC is told by one school district that their average teacher spends 20 hours every year in professional development related to assessment and accountability. Despite how preposterous this number should sound, this one piece of hearsay is used by the NRC to estimate the amount of time all teachers everywhere, whether involved in testing or not, spend annually in related professional development.
• Moreover, professional development related to testing and accountability is assumed to be unrelated to regular instruction and, so, is counted as a completely separate, added-on (i.e., marginal) cost.
• The NRC counts educator time working on standard-setting and other test development panels as “two or three days” which, as anyone who has worked in test development knows, is a high estimate. One to two days is more realistic.
Finally, the NRC studies testing and accountability in only several school districts in only three states. But, according to them, the GAO report which analyzed details from all 48 states involved in testing and over 500 school districts …is the study that left stuff out. In the end, the NRC estimates for testing and accountability costs, are in their own words “about six times higher” than previous estimates.
Incentives and Test-Based Accountability in Education, 2011
“The celebrity professor is a new phenomenon and not a good one. In celebrity-driven academia, ‘getting ahead’ means beating other people, which means establishing a personal reputation and denying it, to the extent possible, to rivals.” — Harry Lewis
By coincidence, a draft copy of this most current National Research Council study was released about the same time I was writing up the results of my decade-long research summary and meta-analyses of the effect of testing on student achievement. So, naturally, I was interested to see how much more the resource-rich NRC could do with the same material. As it turns out, they have remained true to form—this report covers only a highly-selective, tiny fraction of the research literature—but implies that it is all there is to be found—and ignores or declares nonexistent the vast majority of relevant research.
For my work, I looked at studies published in English between 1910 and 2010 that I could obtain and review before a self-imposed deadline in 2010. My coverage of the research literature is far from complete. It includes 244 qualitative studies (e.g., direct observations, site visits, interviews, case studies), 813 individual item-response group combinations from 247 survey studies (e.g., program evaluation surveys, opinion polls), and 640 separate measurements of effects from 177 quantitative research studies (e.g., regression analysis, structural equation modeling, pre-post comparison, experimental design, or interrupted time series design). In total, I analyzed 1,671 separate effects from 668 studies.
The domain of coverage for the NRC study is nominally larger than mine because they purported to analyze non-test incentives and effects on outcomes other than achievement.
Nonetheless, they include but 18 studies in their analysis.
They whittle down the number by eliminating from consideration all qualitative and survey studies and studies conducted prior to the past two decades. For “pre-NCLB” studies from the 1990s to the mid-2000s, the NRC simply expropriates a meta-analysis conducted by Jaekyung Lee (2008) that covered only 14 “cross-state”, hugely large-scale studies, the type least likely to find a strong effect. Unlike small-scale studies and, particularly, experiments that can focus on the factor of interest, empirical studies of large-scale testing programs comprise hundreds of factors for programs with multiple goals and objectives. In statistical lingo, such studies are full of “noise”.
Lee calculated an average effect size across the 14 cross-state studies of 0.08, a positive, but very weak effect. And, that is what the NRC goes with. They conclude that, prior to the date the NCLB Act’s stakes kicked in (about 2006) studies of the effect of testing on student achievement found a 0.08 average effect size.
I include all of Lee’s studies, and his calculations, in my own analysis of quantitative studies that meet the NRC criteria for inclusion, but I also include a multitude of studies the NRC deliberately leaves out. Counting relevant studies only from the same time period—1990 to 2005—the mean effect size is 0.82, ten times larger than the NRC’s. Moreover, that is the simple, “bare-bones” effect size, unadjusted for measurement artifacts—adjustments that would make it even larger.
For the post-NCLB period, the NRC includes seven other large-scale studies that accumulate feeble effect sizes, and 13 studies published between 2002 and 2010 of “incentive experiments using rewards” from India, Israel, Kenya, and the United States.
On the whole, the NRC selection of studies is quite odd, and ridiculously unrepresentative of the research literature it purports to summarize. One consistency in the selection is apparent, however. Only studies finding very small effects are included.
In my review of the 1999 study, High Stakes, I criticized the NRC for restricting its literature survey to U.S. education research, ignoring relevant research conducted in other countries and in other disciplines, such as psychology and economics.
Behold. This newer report mentions some relevant work conducted overseas and by psychologists and economists. Still, the sample is highly selective and excludes the most seminal work in the field. The NRC finds a small group of work that reflects the in-group bias, and the larger world of research and researchers is ignored as if it did not exist.
Most notably, in addition to the usual education researchers, this NRC study covers the work of a crew of young economists, who reach the preferred conclusions (of feeble effect sizes). It turns out that this group of economists shares another characteristic in common with the NRC veterans from CRESST and Boston College—dismissive reviews.
One economist whose work is discussed at length in Incentives and Test-Based Accountability in Education claims to have conducted the first systematic empirical study of teacher cheating (in the early 2000s), the first case study of an urban school district comprehensive accountability system (in 2003), one of the first studies of school-based accountability utilizing individual student data (in 2002), one of the first studies of high-stakes testing (in 2002), and one of the first studies of the effect of grade-promotion testing (in 2002). The same fellow declared (in 2001) there to be no empirical research on minimum competency testing programs or high school graduation exams. The research literature fluffed off by just this one person includes several hundred studies dating back to the 1910s.
Another economist well-regarded by the NRC also declared (in 2002) there to be little to no evidence of the effects of testing or accountability systems. A third economist declared (in 1999) there to be little to no empirical work on school-based incentive programs. A fourth declared (in 1996) "Virtually no evidence exists about the merits or flaws of MCTs [minimum competency tests]". A fifth claimed (in 2005) “there is almost no research on the impact of remediation on student outcomes.” A sixth claimed (in 2000) that a paper he had just written “…provides the first empirical evidence on the effects of grading standards, measured at the teacher level."
Whereas all but a trivial amount of the great mass of relevant research is ignored, the work of NRC study committee members is cited liberally. Daniel Koretz wins the prize for the most citations and references with twelve and nine. Overall, 48 citations and 40 references (out of 200) go to committee members’ work. Over 30 references cite CRESST work. The bulk of the rest cite the work of close friends and colleagues, or earlier NRC studies.
At the same time, a Who’s Who of the leading researchers in the field, past and present, goes missing, names such as: John Hattie, Roddy Roediger, John Bishop, Frank Schmidt, W.J. Haynie, Harold Wenglinsky, Linda Winfield, C.C. Ross, E.H. Jones, Mike McDaniel, Lorin Anderson, J.R. Nation, J.H. Block, Carol Parke, S.F. Stager, Arlen Gullickson, Lynn Fuchs, Douglas Fuchs, Kathy Green, Max Eckstein, Harold Noah, Benjamin Bloom, Jeffrey Karpicke, Michael Beck, Stephen Heynemann, David Driscoll, William D. Schafer, Francine Hultgren, Willis Hawley, James H. McMillan, Elizabeth Marsh, Susan Brookhart, Gene Bottoms, Gordon Cawelti, Lorna Earl, Mike Smoker, David Grissmer, Arthur Powell, Harold Stevenson, Hunter Boylan, Elana Shohamy, Aletta Grisay, Chris Whetton, Steve Ferrara, Glynn Ligon, Micheline Perrin, Thomas Fischer, A. Graham Down, Nigel Brooke, John Oxenham, Caroline Gipps, Arthur Hughes, D. Pennycuick, John Poggio, Anthony Somerset, John O. Anderson, Noel McGinn, Anne Anastasi, Nick Theobald, David Miller, Linda Bond, Nancy Protheroe, Floraline Stevens, Thomas Corcoran, Clement Stone, Suzanne Lane, Frank Dempster, and state agencies in Massachusetts, Florida, and South Carolina.
And, those are just names of some folk who have conducted one or more individual studies. Others have summarized batches of several to many studies in meta-analyses or literature reviews, for example (in chronological order): Panlasigui (1928); Ross (1942); Kirkland (1971); Proger & Mann (1973); Jones (1974); Bjork (1975); Peckham & Roe (1977); Wildemuth (1977); Jackson & Battiste (1978); Kulik, Kulik, Bangert-Drowns, & Schwalb (1983–1991); Natriello & Dornbusch (1984); Dawson & Dawson (1985); Levine (1985); Resnick & Resnick (1985); Guskey & Gates (1986); Hembree (1987); Crooks (1988); Dempster (1991); Adams & Chapman (2002); Locke & Latham (2002); Roediger & Karpicke (2006); and Basol & Johanson (2009).
Long lists of many more relevant names and studies that, in most cases, accumulated results unwanted by CRESST and NRC researchers can be found in Phelps 2003, 2005, 2007, and 2008/2009a.
You will find none of this research and none of these researchers mentioned in Incentives and Test-Based Accountability in Education. Yet, the report claims to summarize the relevant literature. Meanwhile, as in earlier NRC reports, this one declares that important research questions remain unanswered, the implication being that these dismissive reviewers should be given millions of dollars to do the research they have declared nonexistent.
Finally, this NRC report advances its pet theory of “test-score inflation”, while excluding the full abundance of counterevidence, thus recommending exactly the wrong policy to address a very serious and very topical problem (see Phelps, 2011a, 2011b).
In the 1980s, a young medical resident working in a high-poverty region of West Virginia heard local school officials claim that their children scored above the national average on standardized tests. Skeptical, he investigated further and ultimately discovered that every U.S. state administering nationally-normed tests claimed to score above average, a statistical impossibility. The phenomenon was tagged the “Lake Wobegon Effect” after Garrison Keillor’s “News from Lake Wobegon” radio comedy sketch, in which “all the children are above average”.
The West Virginia doctor, John Jacob Cannell, M.D., would move on to New Mexico and, eventually, California, but not before documenting his investigations in two self-published books with titles, “How All Fifty States Are above the National Average” (1987) and “How Public Educators Cheat on Standardized Achievement Tests” (1989).
As usually happens after newsworthy school scandals, policy makers and policy commentators expressed disapproval, wrote opinion pieces, formed committees, and, in due course, forgot about it. Deep dives into the topic were left to professional education researchers, the vast majority of whom worked then, as now, as faculty at graduate schools of education, where they shared a vested interest in defending the status quo.
Dr. Cannell cited educator dishonesty and lax security in test administrations as the primary culprits in the Lake Wobegon Effect, also known as “test score inflation” or “artificial test score gains”. It is easy to understand why. Back then, it was common for states and school districts to purchase nationally-normed standardized tests “off the shelf” and handle all aspects of test administration themselves. Moreover, to reduce costs, it was common to reuse the same test forms (and test items) year after year. Even if educators did not intentionally cheat, over time they became familiar with the test forms and items and could easily prepare their students for them. With test scores rising over time, administrators and elected officials could claim credit for increasing learning.
Test security was so lax because they were diagnostic and monitoring tests that “didn’t count”—only one of the dozens of state tests Cannell examined was both nationally-normed and “high-stakes”—involving direct consequences for the educators or students involved.
Regardless the fact that there were no stakes attached to Cannell’s tests, however, prominent education researchers blamed “high stakes” for the test-score inflation he found (Koretz, et al, 1991; Koretz, 2008). Cannell had exhorted the nation to pay attention to a serious problem of educator dishonesty and lax test security, but education insiders co-opted his discovery and turned it to their own advantage (Staradamskis, 2008; Phelps, 2008/2009b, 2010).
“There are many reasons for the Lake Wobegon Effect, most of which are less sinister than those emphasized by Cannell,” said the co-director of CRESST (Linn, 2000, p. 7). After Dr. Cannell left the debate and went on to practice medicine, this federally-funded education professor and his colleagues would repeat the mantra many times—high stakes, not lax security, cause test-score inflation.
It is most astonishing that they stick with the notion because it is so obviously wrong. The SAT and ACT are tests with stakes—one’s score on either helps determine which college one attends. But, they show no evidence of test-score inflation. (Indeed, the SAT was re-centered in the 1990s because of score deflation.) The most high-stakes tests of all—occupational licensure tests—show no evidence of test-score inflation. Both licensure tests and the SAT and ACT, however, are administered with tight security and ample test form and item rotation.
Spot the causal factor
No test-score inflation
e.g., SAT, ACT, licensure examinations
Test-score inflation possible
e.g., some internally administered district and state examinations
No test-score inflation
e.g., NAEP, other externally administered examinations
Test-score inflation possible
e.g., some internally administered district and state examinations, such as those Cannell investigated