VAMboozled!: Kane is at It, Again: “Statistically Significant” Claims Exaggerated to Influence Policy
In a recent post, I critiqued a fellow academic and value-added model (VAM) supporter — Thomas Kane, an economics professor from Harvard University who also directed the $45 million worth of Measures of Effective Teaching (MET) studies for the Bill & Melinda Gates Foundation. Kane has been the source of multiple posts on this blog (see also here, here, and here) as he is a very public figure, very often backing, albeit often not in non-peer-reviewed technical reports and documents, series of exaggerated, and “research-based” claims. In this prior post, I more specifically critiqued the overstated claims he made in a recent National Public Radio (NPR) interview titled: “There Is No FDA For Education. Maybe There Should Be.”
Well, a colleague recently emailed me another such document authored by Kane (and co-written with four colleagues), titled: “Teaching Higher: Educators’ Perspectives on Common Core Implementation.” While this one is quite methodologically sound (i.e., as assessed via a thorough read of the main text of the document, including all footnotes and appendices), it is Kane’s set of claims, again, that are of concern, especially knowing that this report, even though it too has not yet been externally vetted or reviewed, will likely have a policy impact. The main goal of this report is clearly (although not made explicit) to endorse, promote, and in many ways save the Common Core State Standards (CCSS). I emphasize the word save in that clearly, and especially since the passage of the Every Student Succeeds Act (ESSA), many states have rejected the still highly controversial Common Core. I also should note that researchers in this study clearly conducted this study with similar a priori conclusions in mind (i.e., that the Common Core should be saved/promoted); hence, future peer review of this piece may be out of the question as the bias evident in the sets of findings would certainly be a “methodological issue,” again, likely preventing a peer-reviewed publication (see, for example, the a priori conclusion that “[this] study highlights an important advantage of having a common set of standards and assessments across multiple states,” in the abstract (p. 3).
First I will comment on the findings regarding the Common Core, as related to value-added models (VAMs). Next, I will comment on Section III of the report, about “Which [Common Core] Implementation Strategies Helped Students Succeed?” (p. 17). This is where Kane and colleagues “link[ed] teachers’ survey responses [about the Common Core] to their students’ test scores on the 2014–2015 PARCC [Partnership for Assessment of Readiness for College and Careers] and SBAC [Smarter Balanced Assessment Consortium] assessments [both of which are aligned to the Common Core Standards]… This allowed [Kane et al.] to investigate which strategies and which of the [Common Core-related] supports [teachers] received were associated with their performance on PARCC and SBAC,” controlling for a variety of factors including teachers’ prior value-added (p. 17).
With regards to the Common Core sections, Kane et al. lay claims like: “Despite the additional work, teachers and principals in the five states [that have adopted the Common Core = Delaware, Maryland, Massachusetts, New Mexico, and Nevada] have largely embraced [emphasis added] the new standards” (p. 3). They mention nowhere, however, the mediating set of influences interfering with such a claim, that likely lead to this claim entirely or at least in part – that many teachers across the nation have been forced, by prior federal and current state mandates (e.g., in New Mexico), to “embrace the new standards.” Rather, Kane et al. imply throughout the document that this “embracement” is a sure sign that teachers and principals are literally taking the Common Core into and within their open arms. The same interference is at play with their similar claim that “Teachers in the five study states have made major changes [emphasis in the original] in their lesson plans and instructional materials to meet the CCSS” (p. 3). Compliance is certainly a intervening factor, again, likely contaminating and distorting the validity of both of these claims (which are two of the four total claims highlighted throughout the document’s (p. 3)).
Elsewhere, Kane et al. claim that “The new standards and assessments represent a significant challenge for teachers and students” (p. 6), along with an accompanying figure they use to illustrate how proficiency (i.e., the percent of students labeled as proficient) on these five states’ prior tests has decreased, indicating more rigor or a more “significant challenge for teachers and students” thanks to the Common Core. What they completely ignore again, however, is that the cut scores used to define “proficiency” are arbitrary per state, as was their approach to define “proficiency” across states in comparison (see footnote four). What we also know from years of research on such tests is that whenever a state introduces a “new and improved” test (e.g., the PARCC and SBAC tests), which is typically tied to “new and improved standards” (e.g., the Common Core), lower “proficiency” rates are observed. This has happened countless times across states, and certainly prior to the introduction of the PARCC and SBAC tests. Thereafter, the state typically responds with the same types of claims, that “The new standards and assessments represent a significant challenge for teachers and students.” These claims are meant to signal to the public that at last “we” are holding our teachers and students accountable for their teaching and learning, but thereafter, again, proficiency cut scores are arbitrarily redefined (among other things), and then five or ten years later “new and improved” tests and standards are needed again. In other words, this claim is nothing new and it should not be interpreted as such, but it should rather be interpreted as aligned with Einstein’s definition of insanity (i.e., repeating the same behaviors over and over again in the hopes that different results will ultimately materialize) as this is precisely what we as a nation have been doing since the minimum competency era in the early 1980s.
Otherwise, Kane et al.’s other two claims were related to “Which [Common Core] Implementation Strategies Helped Students Succeed” (p. 17), as mentioned. They assert first that “In mathematics, [they] identified three markers of successful implementation: more professional development days, more classroom observations with explicit feedback tied to the Common Core, and the inclusion of Common Core-aligned student outcomes in teacher evaluations. All were associated with statistically significantly [emphasis added] higher student performance on the PARCC and [SBAC] assessments in mathematics” (p. 3, see also p. 20). They assert second that “In English language arts, [they] did not find evidence for or against any particular implementation strategies” (p. 3, see also p. 20).
What is highly problematic about these claims is that the three correlated implementation strategies noted, again as significantly associated with teachers’ students’ test-based performance on the PARCC and SBAC mathematics assessments, were “statistically significant” (determined by standard p or “probability” values under which findings that may have happened due to chance are numerically specified). But, they were not really practically significant, at all. There IS a difference whereby “statistically significant” findings may not be “practically significant,” or in this case “policy relevant,” at all. While many misinterpret “statistical significance” as an indicator of strength or importance, it is not. Practical significance is.
As per the American Statistical Association’s (ASA) recently released “Statement on P-Values,” statistical significance “is not equivalent to scientific, human, or economic significance…Any effect, no matter how tiny, can produce a small p-value [i.e., “statistical significance”] if the sample size or measurement precision is high enough” (p. 10); hence, one must always check for practical significance when making claims about statistical significance, like Kane et al. actually do here, but do here in a similar inflated vein.
As their Table 6 shows (p. 20), the regression coefficients related to these three areas of “statistically significant” influence on teachers’ students’ test-based performance on the new PARCC and SBAC mathematics tests (i.e., more professional development days, more classroom observations with explicit feedback tied to the Common Core, and the inclusion of Common Core-aligned student outcomes in teacher evaluations) yielded the following coefficients, respectively: 0.045 (p < 0.01), 0.044 (p < 0.05), and 0.054 (p < 0.01). They then use as an example the 0.044 (p < 0.05) coefficient (as related to more classroom observations with explicit feedback tied to the Common Core) and explain that “a difference of one standard deviation in the observation and feedback index was associated with an increase of 0.044 standard deviations in students’ mathematics test scores—roughly the equivalent of 1.4 scale score points on the PARCC assessment and 4.1 scale score points on the SBAC.”
In order to generate sizable and policy relevant improvement in test scores, (e.g., by half of a standard deviation), the observation and feedback index should jump up by 11 standard deviations! In addition, given that scale score points do not equal raw or actual test items (e.g., scale score-to-actual test item relationships are typically in the neighborhood of 4 or 5 scale scores points to 1 actual test item), this likely also means that Kane’s interpretations (i.e., mathematics scores were roughly the equivalent of 1.4 scale score points on the PARCC and 4.1 scale score points on the SBAC) actually mean 1/4th or 1/5th of a test item in mathematics on the PARCC and 4/5th of or 1 test item on the SBAC. This hardly “Provides New Evidence on Strategies Related to Improved Student Performance,” unless you define improved student performance as something as little as 1/5th of a test item.
This is also not what Kane et al. claim to be “a moderately sizeable effect!” (p. 21). These numbers should not even be reported, much less emphasized as policy relevant/significant, unless perhaps equivalent to at least 0.25 standard deviations on the test (as a quasi-standard/accepted minimum). Likewise, the same argument can be made about the other three coefficients derived via these mathematics tests. See also similar claims that they asserted (e.g., that “students perform[ed] better when teachers [were] being evaluated based on student achievement” (p. 21).
Because the abstract (and possibly conclusions) section are the main sections of this paper that are likely to have the most educational/policy impact, especially when people do not read all of the text, footnotes, and abstract contents of this entire document, this is irresponsible, and in many ways contemptible. This is also precisely the reason why, again, Kane’s calls for a Federal Drug Administration (FDA) type of entity for education are also so ironic (as explained in my prior post here).
This blog post has been shared by permission from the author.
Readers wishing to comment on the content are encouraged to do so via the link to the original post.
Find the original post here:
The views expressed by the blogger are not necessarily those of NEPC.