Bias in Teacher Observations, As Well

Following a post last month titled “New Empirical Evidence: Students’ ‘Persistent Economic Disadvantage’ More Likely to Bias Value-Added Estimates,” Matt Barnum — senior staff writer for The 74, an (allegedly) non-partisan, honest, and fact-based news site backed by Editor-in-Chief Campbell Brown and covering America’s education system “in crisis” (see, also, a prior post about The 74 here) — followed up with a tweet via Twitter. He wrote: “Yes, though [bias caused by economic disadvantage] likely applies with equal or even more force to other measures of teacher quality, like observations.” I replied via Twitter that I disagreed with this statement in that I was unaware of research in support of his assertion, and Barnum sent me two articles to review thereafter.

I attempted to review both of these articles herein, although I quickly figured out that I had actually read and reviewed the first (2014) piece on this blog (see original post here, see also a 2014 Brookings Institution article summarizing this piece here). In short, in this study researchers found that the observational components of states’ contemporary teacher systems certainly “add” more “value” than their value-added counterparts, especially for (in)formative purposes. However, researchers  found that observational bias also exists, as akin to value-added bias, whereas teachers who are non-randomly assigned students who enter their classrooms with higher levels of prior achievement tend to get higher observational scores than teachers non-randomly assigned students entering their classrooms with lower levels of prior achievement. Researchers concluded that because districts “do not have processes in place to address the possible biases in observational scores,” statistical adjustments might be made to offset said bias, as might external observers/raters be brought in to yield more “objective” observational assessments of teachers.

For the second study, and this post here, I gave this one a more thorough read (you can find the full study, pre-publication here). Using data from the Measures of Effective
Teaching (MET) Project, in which random assignment was used (or more accurately attempted), researchers also explored the extent to which students enrolled in teachers’ classrooms influence classroom observational scores.

They found, primarily, that:

  1. “[T]he context in which teachers work—most notably, the incoming academic performance of their students—plays a critical role in determining teachers’ performance” as measured by teacher observations. More specifically, “ELA [English/language arts] teachers were more than twice as likely to be rated in the top performance quintile if [nearly randomly] assigned the highest achieving students compared with teachers assigned the low-est achieving students,” and “math teachers were more than 6 times as likely.” In addition, “approximately half of the teachers—48% in ELA and 54% in math—were rated in the top two performance quintiles if assigned the highest performing students, while 37% of ELA and only 18% of math teachers assigned the lowest performing students were highly rated based on classroom observation scores”
  2. “[T]he intentional sorting of teachers to students has a significant influence on measured performance” as well. More specifically, results further suggest that “higher performing students [are, at least sometimes] endogenously sorted into the classes of higher performing teachers…Therefore, the nonrandom and positive assignment of teachers to classes of students based on time-invariant (and unobserved) teacher
    characteristics would reveal more effective teacher performance, as measured by classroom observation scores, than may actually be true.”

So, the non-random assignment of teachers biases both the value-added and observational components written into America’s now “more objective” teacher evaluation systems, as (formerly) required of all states that were to comply with federal initiatives and incentives (e.g., Race to the Top). In addition, when those responsible for assigning students to classrooms (sub)consciously favor teachers with high, prior observational scores, this exacerbates the issues. This is especially important when observational (and value-added) data are to be used for high-stakes accountability systems in that the data yielded via really both measurement systems may be less likely to reflect “true” teaching effectiveness due to “true” bias. “Indeed, teachers working with higher achieving students tend to receive higher performance ratings, above and beyond that which might be attributable to aspects of teacher quality,” and vice-versa.

Citation Study #1: Whitehurst, G. J., Chingos, M. M., & Lindquist, K. M. (2014). Evaluating teachers with classroom observations: Lessons learned in four districts. Washington, DC: Brookings Institution. Retrieved from https://www.brookings.edu/wp-content/uploads/2016/06/Evaluating-Teachers-with-Classroom-Observations.pdf

Citation Study #2: Steinberg, M. P., & Garrett, R. (2016). Classroom composition and measured teacher performance: What do teacher observation scores really measure? Educational Evaluation and Policy Analysis, 38(2), 293-317. doi:10.3102/0162373715616249  Retrieved from http://static.politico.com/58/5f/f14b2b144846a9b3365b8f2b0897/study-of-classroom-observations-of-teachers.pdf

 

One Score and Seven Policy Iterations Ago…

I just read what might be one of the best articles I’ve read in a long time on using test scores to measure teacher effectiveness, and why this is such a bad idea. Not surprisingly, unfortunately, this article was written 20 years ago (i.e., 1986) by – Edward Haertel, National Academy of Education member and recently retired Professor at Stanford University. If the name sounds familiar, it should as Professor Emeritus Haertel is one of the best on the topic of, and history behind VAMs (see prior posts about his related scholarship here, here, and here). To access the full article, please scroll to the reference at the bottom of this post.

Heartel wrote this article when at the time policymakers were, like they still are now, trying to hold teachers accountable for their students’ learning as measured on states’ standardized test scores. Although this article deals with minimum competency tests, which were in policy fashion at the time, about seven policy iterations ago, the contents of the article still have much relevance given where we are today — investing in “new and improved” Common Core tests and still riding on unsinkable beliefs that this is the way to reform the schools that have been in despair and (still) in need of major repair since 20+ years ago.

Here are some of the points I found of most “value:”

  • On isolating teacher effects: “Inferring teacher competence from test scores requires the isolation of teaching effects from other major influences on student test performance,” while “the task is to support an interpretation of student test performance as reflecting teacher competence by providing evidence against plausible rival hypotheses or interpretation.” While “student achievement depends on multiple factors, many of which are out of the teacher’s control,” and many of which cannot and likely never will be able to be “controlled.” In terms of home supports, “students enjoy varying levels of out-of-school support for learning. Not only may parental support and expectations influence student motivation and effort, but some parents may share directly in the task of instruction itself, reading with children, for example, or assisting them with homework.” In terms of school supports, “[s]choolwide learning climate refers to the host of factors that make a school more than a collection of self-contained classrooms. Where the principal is a strong instructional leader; where schoolwide policies on attendance, drug use, and discipline are consistently enforced; where the dominant peer culture is achievement-oriented; and where the school is actively supported by parents and the community.” This, all, makes isolating the teacher effect nearly if not wholly impossible.
  • On the difficulties with defining the teacher effect: “Does it include homework? Does it include self-directed study initiated by the student? How about tutoring by a parent or an older sister or brother? For present purposes, instruction logically refers to whatever the teacher being evaluated is responsible for, but there are degrees of responsibility, and it is often shared. If a teacher informs parents of a student’s learning difficulties and they arrange for private tutoring, is the teacher responsible for the student’s improvement? Suppose the teacher merely gives the student low marks, the student informs her parents, and they arrange for a tutor? Should teachers be credited with inspiring a student’s independent study of school subjects? There is no time to dwell on these difficulties; others lie ahead. Recognizing that some ambiguity remains, it may suffice to define instruction as any learning activity directed by the teacher, including homework….The question also must be confronted of what knowledge counts as achievement. The math teacher who digresses into lectures on beekeeping may be effective in communicating information, but for purposes of teacher evaluation the learning outcomes will not match those of a colleague who sticks to quadratic equations.” Much if not all of this cannot and likely never will be able to be “controlled” or “factored” in or our, as well.
  • On standardized tests: The best of standardized tests will (likely) always be too imperfect and not up to the teacher evaluation task, no matter the extent to which they are pitched as “new and improved.” While it might appear that these “problem[s] could be solved with better tests,” they cannot. Ultimately, all that these tests provide is “a sample of student performance. The inference that this performance reflects educational achievement [not to mention teacher effectiveness] is probabilistic [emphasis added], and is only justified under certain conditions.” Likewise, these tests “measure only a subset of important learning objectives, and if teachers are rated on their students’ attainment of just those outcomes, instruction of unmeasured objectives [is also] slighted.” Like it was then as it still is today, “it has become a commonplace that standardized student achievement tests are ill-suited for teacher evaluation.”
  • On the multiple choice formats of such tests: “[A] multiple-choice item remains a recognition task, in which the problem is to find the best of a small number of predetermined alternatives and the cri- teria for comparing the alternatives are well defined. The nonacademic situations where school learning is ultimately ap- plied rarely present problems in this neat, closed form. Discovery and definition of the problem itself and production of a variety of solutions are called for, not selection among a set of fixed alternatives.”
  • On students and the scores they are to contribute to the teacher evaluation formula: “Students varying in their readiness to profit from instruction are said to differ in aptitude. Not only general cognitive abilities, but relevant prior instruction, motivation, and specific inter- actions of these and other learner characteristics with features of the curriculum and instruction will affect academic growth.” In other words, one cannot simply assume all students will learn or grow at the same rate with the same teacher. Rather, they will learn at different rates given their aptitudes, their “readiness to profit from instruction,” the teachers’ instruction, and sometimes despite the teachers’ instruction or what the teacher teaches.
  • And on the formative nature of such tests, as it was then: “Teachers rarely consult standardized test results except, perhaps, for initial grouping or placement of students, and they believe that the tests are of more value to school or district administrators than to themselves.”

Sound familiar?

Reference: Haertel, E. (1986). The valid use of student performance measures for teacher evaluation. Educational Evaluation and Policy Analysis, 8(1), 45-60.

Another Oldie but Still Very Relevant Goodie, by McCaffrey et al.

I recently re-read an article in full that is now 10 years old, or 10 years out, as published in 2004 and, as per the words of the authors, before VAM approaches were “widely adopted in formal state or district accountability systems.” Unfortunately, I consistently find it interesting, particularly in terms of the research on VAMs, to re-explore/re-discover what we actually knew 10 years ago about VAMs, as most of the time, this serves as a reminder of how things, most of the time, have not changed.

The article, “Models for Value-Added Modeling of Teacher Effects,” is authored by Daniel McCaffrey (Educational Testing Service [ETS] Scientist, and still a “big name” in VAM research), J. R. Lockwood (RAND Corporation Scientists),  Daniel Koretz (Professor at Harvard), Thomas Louis (Professor at Johns Hopkins), and Laura Hamilton (RAND Corporation Scientist).

At the point at which the authors wrote this article, besides the aforementioned data and data base issues, were issues with “multiple measures on the same student and multiple teachers instructing each student” as “[c]lass groupings of students change annually, and students are taught by a different teacher each year.” Authors, more specifically, questioned “whether VAM really does remove the effects of factors such as prior performance and [students’] socio-economic status, and thereby provide[s] a more accurate indicator of teacher effectiveness.”

The assertions they advanced, accordingly and as relevant to these questions, follow:

  • Across different types of VAMs, given different types of approaches to control for some of the above (e.g., bias), teachers’ contribution to total variability in test scores (as per value-added gains) ranged from 3% to 20%. That is, teachers can realistically only be held accountable for 3% to 20% of the variance in test scores using VAMs, while the other 80% to 97% of the variance (stil) comes from influences outside of the teacher’s control. A similar statistic (i.e., 1% to 14%) was similarly and recently highlighted in the recent position statement on VAMs released by the American Statistical Association.
  • Most VAMs focus exclusively on scores from standardized assessments, although I will take this one-step further now, noting that all VAMs now focus exclusively on large-scale standardized tests. This I evidenced in a recent paper I published here: Putting growth and value-added models on the map: A national overview).
  • VAMs introduce bias when missing test scores are not missing completely at random. The missing at random assumption, however, runs across most VAMs because without it, data missingness would be pragmatically insolvable, especially “given the large proportion of missing data in many achievement databases and known differences between students with complete and incomplete test data.” The really only solution here is to use “implicit imputation of values for unobserved gains using the observed scores” which is “followed by estimation of teacher effect[s] using the means of both the imputed and observe gains [together].”
  • Bias “[still] is one of the most difficult issues arising from the use of VAMs to estimate school or teacher effects…[and]…the inclusion of student level covariates is not necessarily the solution to [this] bias.” In other words, “Controlling for student-level covariates alone is not sufficient to remove the effects of [students’] background [or demographic] characteristics.” There is a reason why bias is still such a highly contested issue when it comes to VAMs (see a recent post about this here).
  • All (or now most) commonly-used VAMs assume that teachers’ (and prior teachers’) effects persist undiminished over time. This assumption “is not empirically or theoretically justified,” either, yet it persists.

These authors’ overall conclusion, again from 10 years ago but one that in many ways still stands? VAMs “will often be too imprecise to support some of [its] desired inferences” and uses including, for example, making low- and high-stakes decisions about teacher effects as produced via VAMs. “[O]btaining sufficiently precise estimates of teacher effects to support ranking [and such decisions] is likely to [forever] be a challenge.”

Special Issue of “Educational Researcher” (Paper #9 of 9): Amidst the “Blooming Buzzing Confusion”

Recall that the peer-reviewed journal Educational Researcher (ER) – published a “Special Issue” including nine articles examining value-added measures (VAMs). I have reviewed the last of nine articles (#9 of 9), which is actually a commentary titled “Value Added: A Case Study in the Mismatch Between Education Research and Policy.” This commentary is authored by Stephen Raudenbush – Professor of Sociology and Public Policy Studies at the University of Chicago.

Like with the last two commentaries reviewed here and here, Raudenbush writes of the “Special Issue” that, in this topical area, “[r]esearchers want their work to be used, so we flirt with the idea that value-added research tells us how to improve schooling…[Luckily, perhaps] this volume has some potential to subdue this flirtation” (p. 138).

Raudenbush positions the research covered in this “Special Issue,” as well as the research on teacher evaluation and education in general, as being conducted amidst the “blooming buzzing confusion” (p. 138) surrounding the messy world through which we negotiate life. This is why “specific studies don’t tell us what to do, even if they sometimes have large potential for informing expert judgment” (p. 138).

With that being said, “[t]he hard question is how to integrate the new research on teachers with other important strands of research [e.g., effective schools research] in order to inform rather than distort practical judgment” (p. 138). Echoing Susan Moore Johnson’s sentiments, reviewed as article #6 here, this is appropriately hard if we are to augment versus undermine “our capacity to mobilize the “social capital” of the school to strengthen the human capital of the teacher” (p. 138).

On this note, and “[i]n sum, recent research on value added tells us that, by using data from student perceptions, classroom observations, and test score growth, we can obtain credible evidence [albeit weakly related evidence, referring to the Bill & Melinda Gates Foundation’s MET studies] of the relative effectiveness of a set of teachers who teach similar kids [emphasis added] under similar conditions [emphasis added]…[Although] if a district administrator uses data like that collected in MET, we can anticipate that an attempt to classify teachers for personnel decisions will be characterized by intolerably high error rates [emphasis added]. And because districts can collect very limited information, a reliance on district-level data collection systems will [also] likely generate…distorted behavior[s]..in which teachers attempt to “game” the
comparatively simple indicators,” or system (p. 138-139).

Accordingly, “[a]n effective school will likely be characterized by effective ‘distributed’ leadership, meaning that expert teachers share responsibility for classroom observation, feedback, and frequent formative assessments of student learning. Intensive professional development combined with classroom follow-up generates evidence about teacher learning and teacher improvement. Such local data collection efforts [also] have some potential to gain credibility among teachers, a virtue that seems too often absent” (p. 140).

This, might be at least a significant part of the solution.

“If the school is potentially rich in information about teacher effectiveness and teacher improvement, it seems to follow that key personnel decisions should be located firmly at the school level..This sense of collective efficacy [accordingly] seems to be a key feature of…highly effective schools” (p. 140).

*****

If interested, see the Review of Article #1 – the introduction to the special issue here; see the Review of Article #2 – on VAMs’ measurement errors, issues with retroactive revisions, and (more) problems with using standardized tests in VAMs here; see the Review of Article #3 – on VAMs’ potentials here; see the Review of Article #4 – on observational systems’ potentials here; see the Review of Article #5 – on teachers’ perceptions of observations and student growth here; see the Review of Article (Essay) #6 – on VAMs as tools for “egg-crate” schools here; see the Review of Article (Commentary) #7 – on VAMs situated in their appropriate ecologies here; and see the Review of Article #8, Part I – on a more research-based assessment of VAMs’ potentials here and Part II on “a modest solution” provided to us by Linda Darling-Hammond here.

Article #9 Reference: Raudenbush, S. W. (2015). Value added: A case study in the mismatch between education research and policy. Educational Researcher, 44(2), 138-141. doi:10.3102/0013189X15575345

 

 

 

Special Issue of “Educational Researcher” (Paper #8 of 9, Part I): A More Research-Based Assessment of VAMs’ Potentials

Recall that the peer-reviewed journal Educational Researcher (ER) – published a “Special Issue” including nine articles examining value-added measures (VAMs). I have reviewed the next of nine articles (#8 of 9), which is actually a commentary titled “Can Value-Added Add Value to Teacher Evaluation?” This commentary is authored by Linda Darling-Hammond – Professor of Education, Emeritus, at Stanford University.

Like with the last commentary reviewed here, Darling-Hammond reviews some of the key points taken from the five feature articles in the aforementioned “Special Issue.” More specifically, though, Darling-Hammond “reflect[s] on [these five] articles’ findings in light of other work in this field, and [she] offer[s her own] thoughts about whether and how VAMs may add value to teacher evaluation” (p. 132).

She starts her commentary with VAMs “in theory,” in that VAMs COULD accurately identify teachers’ contributions to student learning and achievement IF (and this is a big IF) the following three conditions were met: (1) “student learning is well-measured by tests that reflect valuable learning and the actual achievement of individual students along a vertical scale representing the full range of possible achievement measures in equal interval units” (2) “students are randomly assigned to teachers within and across schools—or, conceptualized another way, the learning conditions and traits of the group of students assigned to one teacher do not vary substantially from those assigned to another;” and (3) “individual teachers are the only contributors to students’ learning over the period of time used for measuring gains” (p. 132).

None of things are actual true (or near to true, nor will they likely ever be true) in educational practice, however. Hence, the errors we continue to observe that continue to prevent VAM use for their intended utilities, even with the sophisticated statistics meant to mitigate errors and account for the above-mentioned, let’s call them, “less than ideal” conditions.

Other pervasive and perpetual issues surrounding VAMs as highlighted by Darling-Hammond, per each of the three categories above, pertain to (1) the tests used to measure value-added is that the tests are very narrow, focus on lower level skills, and are manipulable. These tests in their current form cannot effectively measure the learning gains of a large share of students who are above or below grade level given a lack of sufficient coverage and stretch. As per Haertel (2013, as cited in Darling-Hammond’s commentary), this “translates into bias against those teachers working with the lowest-performing or the highest-performing classes’…and “those who teach in tracked school settings.” It is also important to note here that the new tests created by the Partnership for Assessing Readiness for College and Careers (PARCC) and Smarter Balanced, multistate consortia “will not remedy this problem…Even though they will report students’ scores on a vertical scale, they will not be able to measure accurately the achievement or learning of students who started out below or above grade level” (p.133).

With respect to (2) above, on the equivalence (or rather non-equivalence) of groups of student across teachers classrooms who are the ones whose VAM scores are relativistically compared, the main issue here is that “the U.S. education system is the one of most segregated and unequal in the industrialized world…[likewise]…[t]he country’s extraordinarily high rates of childhood poverty, homelessness, and food insecurity are not randomly distributed across communities…[Add] the extensive practice of tracking to the mix, and it is clear that the assumption of equivalence among classrooms is far from reality” (p. 133). Whether sophisticated statistics can control for all of this variation is one of most debated issues surrounding VAMs and their levels of outcome bias, accordingly.

And as per (3) above, “we know from decades of educational research that many things matter for student achievement aside from the individual teacher a student has at a moment in time for a given subject area. A partial list includes the following [that are also supposed to be statistically controlled for in most VAMs, but are also clearly not controlled for effectively enough, if even possible]: (a) school factors such as class sizes, curriculum choices, instructional time, availability of specialists, tutors, books, computers, science labs, and other resources; (b) prior teachers and schooling, as well as other current teachers—and the opportunities for professional learning and collaborative planning among them; (c) peer culture and achievement; (d) differential summer learning gains and losses; (e) home factors, such as parents’ ability to help with homework, food and housing security, and physical and mental support or abuse; and (e) individual student needs, health, and attendance” (p. 133).

“Given all of these influences on [student] learning [and achievement], it is not surprising that variation among teachers accounts for only a tiny share of variation in achievement, typically estimated at under 10%” (see, for example, highlights from the American Statistical Association’s (ASA’s) Position Statement on VAMs here). “Suffice it to say [these issues]…pose considerable challenges to deriving accurate estimates of teacher effects…[A]s the ASA suggests, these challenges may have unintended negative effects on overall educational quality” (p. 133). “Most worrisome [for example] are [the] studies suggesting that teachers’ ratings are heavily influenced [i.e., biased] by the students they teach even after statistical models have tried to control for these influences” (p. 135).

Other “considerable challenges” include: VAM output are grossly unstable given the swings and variations observed in teacher classifications across time, and VAM output are “notoriously imprecise” (p. 133) given the other errors observed as caused, for example, by varying class sizes (e.g., Sean Corcoran (2010) documented with New York City data that the “true” effectiveness of a teacher ranked in the 43rd percentile could have had a range of possible scores from the 15th to the 71st percentile, qualifying as “below average,” “average,” or close to “above average). In addition, practitioners including administrators and teachers are skeptical of these systems, and their (appropriate) skepticisms are impacting the extent to which they use and value their value-added data, noting that they value their observational data (and the professional discussions surrounding them) much more. Also important is that another likely unintended effect exists (i.e., citing Susan Moore Johnson’s essay here) when statisticians’ efforts to parse out learning to calculate individual teachers’ value-added causes “teachers to hunker down and focus only on their own students, rather than working collegially to address student needs and solve collective problems” (p. 134). Related, “the technology of VAM ranks teachers against each other relative to the gains they appear to produce for students, [hence] one teacher’s gain is another’s loss, thus creating disincentives for collaborative work” (p. 135). This is what Susan Moore Johnson termed the egg-crate model, or rather the egg-crate effects.

Darling-Hammond’s conclusions are that VAMs have “been prematurely thrust into policy contexts that have made it more the subject of advocacy than of careful analysis that shapes its use. There is [good] reason to be skeptical that the current prescriptions for using VAMs can ever succeed in measuring teaching contributions well (p. 135).

Darling-Hammond also “adds value” in one whole section (highlighted in another post forthcoming here), offering a very sound set of solutions, using VAMs for teacher evaluations or not. Given it’s rare in this area of research we can focus on actual solutions, this section is a must read. If you don’t want to wait for the next post, read Darling-Hammond’s “Modest Proposal” (p. 135-136) within her larger article here.

In the end, Darling-Hammond writes that, “Trying to fix VAMs is rather like pushing on a balloon: The effort to correct one problem often creates another one that pops out somewhere else” (p. 135).

*****

If interested, see the Review of Article #1 – the introduction to the special issue here; see the Review of Article #2 – on VAMs’ measurement errors, issues with retroactive revisions, and (more) problems with using standardized tests in VAMs here; see the Review of Article #3 – on VAMs’ potentials here; see the Review of Article #4 – on observational systems’ potentials here; see the Review of Article #5 – on teachers’ perceptions of observations and student growth here; see the Review of Article (Essay) #6 – on VAMs as tools for “egg-crate” schools here; and see the Review of Article (Commentary) #7 – on VAMs situated in their appropriate ecologies here.

Article #8, Part I Reference: Darling-Hammond, L. (2015). Can value-added add value to teacher evaluation? Educational Researcher, 44(2), 132-137. doi:10.3102/0013189X15575346

Houston Lawsuit Update, with Summary of Expert Witnesses’ Findings about the EVAAS

Recall from a prior post that a set of teachers in the Houston Independent School District (HISD), with the support of the Houston Federation of Teachers (HFT) are taking their district to federal court to fight for their rights as professionals, and how their value-added scores, derived via the Education Value-Added Assessment System (EVAAS), have allegedly violated them. The case, Houston Federation of Teachers, et al. v. Houston ISD, is to officially begin in court early this summer.

More specifically, the teachers are arguing that EVAAS output are inaccurate, the EVAAS is unfair, that teachers are being evaluated via the EVAAS using tests that do not match the curriculum they are to teach, that the EVAAS system fails to control for student-level factors that impact how well teachers perform but that are outside of teachers’ control (e.g., parental effects), that the EVAAS is incomprehensible and hence very difficult if not impossible to actually use to improve upon their instruction (i.e., actionable), and, accordingly, that teachers’ due process rights are being violated because teachers do not have adequate opportunities to change as a results of their EVAAS results.

The EVAAS is the one value-added model (VAM) on which I’ve conducted most of my research, also in this district (see, for example, here, here, here, and here); hence, I along with Jesse Rothstein – Professor of Public Policy and Economics at the University of California – Berkeley, who also conducts extensive research on VAMs – are serving as the expert witnesses in this case.

What was recently released regarding this case is a summary of the contents of our affidavits, as interpreted by authors of the attached “EVAAS Litigation UPdate,” in which the authors declare, with our and others’ research in support, that “Studies Declare EVAAS ‘Flawed, Invalid and Unreliable.” Here are the twelve key highlights, again, as summarized by the authors of this report and re-summarized, by me, below:

  1. Large-scale standardized tests have never been validated for their current uses. In other words, as per my affidavit, “VAM-based information is based upon large-scale achievement tests that have been developed to assess levels of student achievement, but not levels of growth in student achievement over time, and not levels of growth in student achievement over time that can be attributed back to students’ teachers, to capture the teachers’ [purportedly] causal effects on growth in student achievement over time.”
  2. The EVAAS produces different results from another VAM. When, for this case, Rothstein constructed and ran an alternative, albeit sophisticated VAM using data from HISD both times, he found that results “yielded quite different rankings and scores.” This should not happen if these models are indeed yielding indicators of truth, or true levels of teacher effectiveness from which valid interpretations and assertions can be made.
  3. EVAAS scores are highly volatile from one year to the next. Rothstein, when running the actual data, found that while “[a]ll VAMs are volatile…EVAAS growth indexes and effectiveness categorizations are particularly volatile due to the EVAAS model’s failure to adequately account for unaccounted-for variation in classroom achievement.” In addition, volatility is “particularly high in grades 3 and 4, where students have relatively few[er] prior [test] scores available at the time at which the EVAAS scores are first computed.”
  4. EVAAS overstates the precision of teachers’ estimated impacts on growth. As per Rothstein, “This leads EVAAS to too often indicate that teachers are statistically distinguishable from the average…when a correct calculation would indicate that these teachers are not statistically distinguishable from the average.”
  5. Teachers of English Language Learners (ELLs) and “highly mobile” students are substantially less likely to demonstrate added value, as per the EVAAS, and likely most/all other VAMs. This, what we term as “bias,” makes it “impossible to know whether this is because ELL teachers [and teachers of highly mobile students] are, in fact, less effective than non-ELL teachers [and teachers of less mobile students] in HISD, or whether it is because the EVAAS VAM is biased against ELL [and these other] teachers.”
  6. The number of students each teacher teaches (i.e., class size) also biases teachers’ value-added scores. As per Rothstein, “teachers with few linked students—either because they teach small classes or because many of the students in their classes cannot be used for EVAAS calculations—are overwhelmingly [emphasis added] likely to be assigned to the middle effectiveness category under EVAAS (labeled “no detectable difference [from average], and average effectiveness”) than are teachers with more linked students.”
  7. Ceiling effects are certainly an issue. Rothstein found that in some grades and subjects, “teachers whose students have unusually high prior year scores are very unlikely to earn high EVAAS scores, suggesting that ‘ceiling effects‘ in the tests are certainly relevant factors.” While EVAAS and HISD have previously acknowledged such problems with ceiling effects, they apparently believe these effects are being mediated with the new and improved tests recently adopted throughout the state of Texas. Rothstein, however, found that these effects persist even given the new and improved.
  8. There are major validity issues with “artificial conflation.” This is a term I recently coined to represent what is happening in Houston, and elsewhere (e.g., Tennessee), when district leaders (e.g., superintendents) mandate or force principals and other teacher effectiveness appraisers or evaluators, for example, to align their observational ratings of teachers’ effectiveness with value-added scores, with the latter being the “objective measure” around which all else should revolve, or align; hence, the conflation of the one to match the other, even if entirely invalid. As per my affidavit, “[t]o purposefully and systematically endorse the engineering and distortion of the perceptible ‘subjective’ indicator, using the perceptibly ‘objective’ indicator as a keystone of truth and consequence, is more than arbitrary, capricious, and remiss…not to mention in violation of the educational measurement field’s Standards for Educational and Psychological Testing” (American Educational Research Association (AERA), American Psychological Association (APA), National Council on Measurement in Education (NCME), 2014).
  9. Teaching-to-the-test is of perpetual concern. Both Rothstein and I, independently, noted concerns about how “VAM ratings reward teachers who teach to the end-of-year test [more than] equally effective teachers who focus their efforts on other forms of learning that may be more important.”
  10. HISD is not adequately monitoring the EVAAS system. According to HISD, EVAAS modelers keep the details of their model secret, even from them and even though they are paying an estimated $500K per year for district teachers’ EVAAS estimates. “During litigation, HISD has admitted that it has not performed or paid any contractor to perform any type of verification, analysis, or audit of the EVAAS scores. This violates the technical standards for use of VAM that AERA specifies, which provide that if a school district like HISD is going to use VAM, it is responsible for ‘conducting the ongoing evaluation of both intended and unintended consequences’ and that ‘monitoring should be of sufficient scope and extent to provide evidence to document the technical quality of the VAM application and the validity of its use’ (AERA Statement, 2015).
  11. EVAAS lacks transparency. AERA emphasizes the importance of transparency with respect to VAM uses. For example, as per the AERA Council who wrote the aforementioned AERA Statement, “when performance levels are established for the purpose of evaluative decisions, the methods used, as well as the classification accuracy, should be documented and reported” (AERA Statement, 2015). However, and in contrast to meeting AERA’s requirements for transparency, in this district and elsewhere, as per my affidavit, the “EVAAS is still more popularly recognized as the ‘black box’ value-added system.”
  12. Related, teachers lack opportunities to verify their own scores. This part is really interesting. “As part of this litigation, and under a very strict protective order that was negotiated over many months with SAS [i.e., SAS Institute Inc. which markets and delivers its EVAAS system], Dr. Rothstein was allowed to view SAS’ computer program code on a laptop computer in the SAS lawyer’s office in San Francisco, something that certainly no HISD teacher has ever been allowed to do. Even with the access provided to Dr. Rothstein, and even with his expertise and knowledge of value-added modeling, [however] he was still not able to reproduce the EVAAS calculations so that they could be verified.”Dr. Rothstein added, “[t]he complexity and interdependency of EVAAS also presents a barrier to understanding how a teacher’s data translated into her EVAAS score. Each teacher’s EVAAS calculation depends not only on her students, but also on all other students with- in HISD (and, in some grades and years, on all other students in the state), and is computed using a complex series of programs that are the proprietary business secrets of SAS Incorporated. As part of my efforts to assess the validity of EVAAS as a measure of teacher effectiveness, I attempted to reproduce EVAAS calculations. I was unable to reproduce EVAAS, however, as the information provided by HISD about the EVAAS model was far from sufficient.”

Economists Declare Victory for VAMs

On a popular economics site, fivethirtyeight.com, authors use “hard numbers” to tell compelling stories, and this time the compelling story told is about value-added models and all of the wonders, thanks to the “hard numbers” derived via model output, they are working to reform the way “we” evaluate and hold teachers accountable for their effects.

In an article titled “The Science Of Grading Teachers Gets High Marks,” this site’s “quantitative editor” (?!?) – Andrew Flowers – writes about how “the science” behind using “hard numbers” to evaluate teachers’ effects is, fortunately for America and thanks to the efforts of (many/most) econometricians, gaining much-needed momentum.

Not to really anyone’s surprise, the featured economics study of this post is…wait for it…the Chetty et al. study at focus of much controversy and many prior posts on this blog (see for example here, here, here, and here). This is the study cited in President Obama’ 2012 State of the Union address when he said that, “We know a good teacher can increase the lifetime income of a classroom by over $250,000,” and this study was more recently the focus of attention when the judge in Vergara v. California cited Chetty et al.’s study as providing evidence that “a single year in a classroom with a grossly ineffective teacher costs students $1.4 million in lifetime earnings per classroom.”

These are the “hard numbers” that have since been duly critiqued by scholars from California to New York since (see, for example, here, here, here, and here), but that’s not mentioned in this post. What is mentioned, however, is the notable work of economist Jesse Rothstein, whose work I have also cited in prior posts (see, for example, here, here, here, and here,), as he has also countered Chetty et al.’s claims, not to mention added critical research to the topic on VAM-based bias.

What is also mentioned, not to really anyone’s surprise again, though, is that Thomas Kane – a colleague of Chetty’s at Harvard who has also been the source of prior VAMboozled! posts (see, for example, here, here, and here), who also replicated Chetty’s results as notably cited/used during the Vergara v. California case last summer, endorses Chetty’s work throughout this same article. Article author “reached out” to Kane “to get more perspective,” although I, for one, question how random this implied casual reach really was… Recall a recent post about our “(Unfortunate) List of VAMboozlers?” Two of our five total honorees include Chetty and Kane – the same two “hard number” economists prominently featured in this piece.

Nonetheless, this article’s “quantitative editor” (?!?) Flowers sides with them (i.e., Chetty and Kane), and ultimately declares victory for VAMs, writing that VAMs ultimately and “accurately isolate a teacher’s impact on students”…”[t]he implication[s] being, school administrators can legitimately use value-added scores to hire, fire and otherwise evaluate teacher performance.”

This “cutting-edge science,” as per a quote taken from Chetty’s co-author Friedman (Brown University), captures it all: “It’s almost like we’re doing real, hard science…Well, almost. But by the standards of empirical social science — with all its limitations in experimental design, imperfect data, and the hard-to-capture behavior of individuals — it’s still impressive….[F]or what has been called the “credibility revolution” in empirical economics, it’s a win.”

New “Causal” Evidence in Support of VAMs

No surprise, really, but 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, is publicly writing in support of VAMs, again. His newest article was recently published on the website of the Brookings Institution, titled “Do Value-Added Estimates Identify Causal Effects of Teachers and Schools?” Not surprisingly as a VAM advocate, in this piece he continues to advance a series of false claims about the wonderful potentials of VAMs, in this case as also yielding causal estimates whereas teachers can be seen as directly causing the growth measured by VAMs (see prior posts about Kane’s very public perspectives here and here).

The best part of this article is where Kane (potentially) seriously considers whether “the short list of control variables captured in educational data systems—prior achievement, student demographics, English language learner [ELL] status, eligibility for federally subsidized meals or programs for gifted and special education students—include the relevant factors by which students are sorted to teachers and schools” and, hence, work to control for bias as these are some of the factors that do (as per the research evidence) distort value-added scores.

The potential surrounding the exploration of this argument, however, quickly turns south as thereafter Kane pontificates using pure logic that “it is possible that school data systems contain the very data that teachers or principals are using to assign students to teachers.” In other words, he painfully attempts to assert his non-research-based argument that as long as principals and teachers use the aforementioned variables to sort students into classrooms, then controlling for said variables should indeed control for the biasing effects caused by the non-random assortment of students into classrooms (and teachers into classrooms, although he does not address that component either).

In addition, he asserts that, “[o]f course, there are many other unmeasured factors [or variables] influencing student achievement—such as student motivation or parental engagement [that cannot or cannot easily be observed]. But as long as those factors are also invisible [emphasis added] to those making teacher and program assignment decisions, our [i.e., VAM statisticians’] inability to control for them” more or less makes not controlling for these other variables inconsequential. In other words, in this article Kane asserts as long as the “other things” principals and teachers use to non-randomly place students into classrooms are “invisible” to the principals and teachers making student placement decisions, these “other things” should not have to be statistically controlled, or factored out. We should otherwise be good to go given the aforementioned variables already observable and available.

As evidenced in a study I wrote with one of my current doctoral students that was recently published in the esteemed, peer-reviewed American Educational Research Journal on this very topic (see the full study here), we set out to better determine how and whether the controls used by value-added researchers to eliminate bias might be sufficient given what indeed occurs in practice when students are placed into classrooms.

We found that both teachers and parents play a prodigious role in the student placement process, in almost nine out of ten schools (i.e., 90% of the time). Teachers and parents (although parents are also not mentioned in Kane’s article) provide both appreciated and sometimes unwelcome insights, regarding what teachers and parents perceive to be the best learning environments for their students or children, respectively. Their added insights typically revolve around, in the following order, students’ in-school behaviors, attitudes, and disciplinary records; students’ learning styles and students’ learning styles as matched with teachers’ teaching styles; students’ personalities and students personalities as matched with teachers’ personalities; students’ interactions with their peers and prior teachers; general teacher types (e.g. teachers who manage their classrooms in perceptibly better ways); and whether students had siblings in potential teachers’ classrooms prior.

These “other things” are not typically if ever controlled for given current VAMs, nor will they likely ever be. In addition, these factors serve as legitimate reasons for class changes during the school year, although whether this, too, is or could be captured in VAMs is highly tentative at best. Otherwise, namely prior academic achievement, special education needs, giftedness, and gender also influence placement decisions. These are variables for which most current VAMs account or control, presumably effectively.

Kane, like other VAM statisticians, tend to (and in many ways have to if they are to continue with their VAM work, despite “the issues”) (over)simplify the serious complexities that come about when random assignment of students to classrooms (and teachers to classrooms) is neither feasible, nor realistic, or outright opposed (as was also clearly evidenced in the above article by 98% of educators, see again here).

The random assignment of students to classrooms (and teachers to classrooms) very rarely happens. Rather, the use of many observable and unobservable variables are used to make such classroom placement decisions, and these variables go well beyond whether students are eligible for free-and-reduced lunches or are English-language learners.

If only the real world surrounding our schools, and in particular the measurement and evaluation of our schools and teachers within them, was so simple and straightforward as Kane and others continue to assume and argue, although much of the time without evidence other than his own or that of his colleagues at Harvard (i.e., 8/17; 47% of the articles cited in Kane’s piece). See also a recent post about this here. In this case, much published research evidence exists to clearly counter this logic and the many related claims herein (see also the other research not cited in this piece but cited in the study highlighted above and linked to again here).

Harvard Economist Deming on VAM-Based Bias

David Deming – an Associate Professor of Education and Economics at Harvard – just published in the esteemed American Economic Review an article about VAM-based bias, in this case when VAMs are used to measure school versus teacher level effects.

Deming appropriately situated his study within the prior works on this topic, including the key works of Thomas Kane (Education and Economics at Harvard) and Raj Chetty (Economics at Harvard). These two, most notably, continue to advance assertions that using students’ prior test scores and other covariates (i.e., to statistically control for students’ demographic/background factors) minimizes VAM-based bias to negligible levels. Deming also situated his study given the notable works of Jesse Rothstein (Public Policy and Economics at the University of California, Berkeley) who continues to evidence VAM-based bias really does exist. The research of these three key players, along with their scholarly disagreements, have also been highlighted in prior posts about VAM-based bias on this blog (see, for example, here and here).

In this study to test for bias, though, Deming used data from Charlotte-Mecklenburg, North Carolina, given a data set derived from a district in which there was quasi-random assignment of students to schools (given a school choice initiative). With these data, Deming tested whether VAM-biased bias was evident across a variety of common VAM approaches, from the least sophisticated VAM (e.g., one year of prior test scores and no other covariates) to the most (e.g., two or more years of prior test score data plus various covariates).

Overall, Deming failed to reject the hypothesis that school-level effects as measured using VAMs are unbiased, almost regardless of the VAM being used. In more straightforward terms, Deming found that school effects as measured using VAMs were rarely if ever biased when compared to his randomized samples. Hence, this work falls inline with prior works countering that bias really does exist (Note: this is a correction from the prior post).

There are still, however, at least three reasons that could lead to bias in either direction (I.e., positive, in favor of school effects or negative, underestimating school effects):

  • VAMs may be biased due to the non-random sorting of students into schools (and classrooms) “on unobserved determinants of achievement” (see also the work of Rothstein, here and here).
  • If “true” school effects vary over time (independent of error), then test-based forecasts based on prior cohorts’ test scores (as is common when measuring the difference between predictions and “actual” growth, when calculating value-added) may be poor predictors of future effectiveness.
  • When students self-select into schools, the impact of attending a school may be different for students who self-select in than for students who do not. The same thing likely holds true for classroom assignment practices, although that is my extrapolation, not Deming’s.

In addition, and in Deming’s overall conclusions that also pertain here, “many other important outcomes of schooling are not measured here. Schools and teachers [who] are good at increasing student achievement may or may not be effective along other important dimensions” (see also here).

For all of these reasons, “we should be cautious before moving toward policies that hold schools accountable for improving their ‘value added” given bias.

Can VAMs Be Trusted?

In a recent paper published in the peer-reviewed journal Education Finance and Policy, coauthors Cassandra Guarino (Indiana University – Bloomington), Mark Reckase (Michigan State University), and Jeffrey Wooldridge (Michigan State University) ask and then answer the following question: “Can Value-Added Measures of Teacher Performance Be Trusted?” While what I write below is taken from what I read via the official publication, I link here to the working paper that was published online via the Education Policy Center at Michigan State University (i.e., not for a fee).

From the abstract, authors “investigate whether commonly used value-added estimation strategies produce accurate estimates of teacher effects under a variety of scenarios. [They] estimate teacher effects [using] simulated student achievement data sets that mimic plausible types of student grouping and teacher assignment scenarios. [They] find that no one method accurately captures true teacher effects in all scenarios, and the potential for misclassifying teachers as high- or low-performing can be substantial.”

From elsewhere in more specific terms, the authors use simulated data to “represent controlled conditions” to most closely match “the relatively simple conceptual model upon which value-added estimation strategies are based.” This is the strength of this research study in that authors’ findings represent best-case scenarios, while when working with real-world and real-life data “conditions are [much] more complex.” Hence, working with various statistical estimators, controls, approaches, and the like using simulated data becomes “the best way to discover fundamental flaws and differences among them when they should be expected to perform at their best.”

They found…

  • “No one [value-added] estimator performs well under all plausible circumstances, but some are more robust than others…[some] fare better than expected…[and] some of the most popular methods are neither the most robust nor ideal.” In other words, calculating value-added regardless of the sophistication of the statistical specifications and controls used is messy, and this messiness can seriously throw off the validity of the inferences to be drawn about teachers, even given the fanciest models and methodological approaches we currently have going (i.e., those models and model specifications being advanced via policy).
  • “[S]ubstantial proportions of teachers can be misclassified as ‘below average’ or ‘above average’ as well as in the bottom and top quintiles of the teacher quality distribution, even in [these] best-case scenarios.” This means that the misclassification errors we are seeing with real-world data, we are also seeing with simulated data. This leads us to even more concern about whether VAMs will ever be able to get it right, or in this case, counter the effects of the nonrandom assignment of students to classrooms and teachers to the same.
  • Researchers found that “even in the best scenarios and under the simplistic and idealized conditions imposed by [their] data-generating process, the potential for misclassifying above-average teachers as below average or for misidentifying the “worst” or “best” teachers remains nontrivial, particularly if teacher effects are relatively small. Applying the [most] commonly used [value-added approaches] results in misclassification rates that range from at least 7 percent to more than 60 percent, depending upon the estimator and scenario.” So even with a pretty perfect dataset, or a dataset much cleaner than those that come from actual children and their test scores in real schools, misclassification errors can impact teachers upwards of 60% of the time.

In sum, researchers conclude that while certain VAMs hold more promise than others, they may not be capable of overcoming the many obstacles presented by the non-random assignment of students to teachers (and teachers to classrooms).

In their own words, “it is clear that every estimator has an Achilles heel (or more than one area of potential weakness)” that can distort teacher-level output in highly consequential ways. Hence, “[t]he degree of error in [VAM] estimates…may make them less trustworthy for the specific purpose of evaluating individual teachers” than we might think.